 Today we're going to close the loop a bit and we're going to come back to free energy and so remember when this all started we started looking at first simple Interactions and biomolecules and then we use those interactions to gradually introduce the concepts of entropy and then we start talking a little bit about free energy and With that free energy we could then start to say understand the hydrophobic effect the hydrogen bonds But we're still talking about the fundamental interactions there, right and then based on that We took another step and started looking at a real biomolecules that we then More to a smaller or greater extent could understand with the free energy concepts And then we even introduced if you're going to model real biomolecules and simulations or something else You will probably start doing you will do that a little bit the last two labs here in particular relate to free energy But that is not just a one-way street So you can also use this type of relatively advanced Simulations calculations to calculate free and it is not just for simple things like a hydrogen bond But for more complicated things such as binding a molecule and we touched upon that a little bit yesterday In drug action more than a little bit in drug design But in drug design what we have done the last 20 years in this field is to do this fast fast fast Because if we can't do it fast enough Well, all right, we had to do it fast because we were not that accurate So we had to somehow compensate accuracy with throughput instead so that we could use this to just screen databases and Based on that screening we then hope to find something and What I'm gonna spend the first half of the lecture today talking about is really the next generation stuff That's coming what we want to do computers are now fast enough that we can do this fairly accurately in the simulations And then I'm gonna spend the second half talking a little bit about our research and the point is It's not just that I'm a corny professor and like to talk about my own research But to partly give you a feeling about research products that go on both in our team and others at PsyLifeLab And I simply for obvious reasons I know more about my own research than others And then this afternoon we have recruited a bunch of students from our labs So that we're gonna split you into groups at PsyLifeLab and have them talk to you during an hour or two depending on what you're interested in But before that we had the other task for you That you're gonna talk about drug design and I'm not gonna ask you a single question, but you can ask me questions If you are completely quiet this can be a very awkward 30 minutes So shoot what did you find interesting unclear Yep So it depends a bit Historically you would always narrow it down with pharmacophores But that assumes that you know something about the binding site right and as computations have gotten better and better We are increasingly trying to do. I was about to say blind docking. It's not quite blind because I know that this is the likely binding site But I have no f-ing idea what will bind there So just screen everything and the point there is that screening through 100 million compounds I can do it dock in docking it might take a week or so And then based on that there is this concept that I didn't tell you about yesterday, but it's called it's important called refinement factor and Let's see here So if you start out with a very large database, let's say 1 million compounds And we know based on pure statistics that Out of that set, let's say that there are it's the probability of having a being a good binder Let's say for argument's sake that it's in the ballpark of 10 to the minus 4. It's completely arbitrary numbers So that means that's 10 that one in 10,000 right? So that means that in theory there should be roughly 100 binders in that binding sets So if I were to pick 1,000 of those molecules and go into the lab the likelihood that I will have again purely random You think the likelihood that I will find a binder is only one in 10 That's not very good And then we might do a docking and what you case it what you think about in terms of docking is how much better are these chances and Normally the docking refinement be a fact might be a factor of five or ten or so But let's for argument's sake say that I have a refact refinement factor of a hundred hundred X refinement So what that would mean that after docking the likelihood in the high ranking compounds that I select The probability would now be a factor hundred better than random So that out of the say the top 1,000 docking molecules if it was completely randomly selected It would be 10 to the minus 4 With this refinement factor It's still 10 to the minus 2 so there were no but the refinement say that sorry the likelihood of a random if I have a compound here is Being a good binder is 10 to the minus 4 think of docking like a black box and This black box selects as another set of molecules, but of course it's not doing it randomly But suddenly after this black box you have improved this probability by a factor of 100 So that the likelihood of a binder here being good is still just 1% which is lousy But the point is that if I now select 1,000 molecules and go into the lab 1,000 times 1% it's going to be I expect to see an average 10 binders So the point is that the actual prediction power of the docking is still lousy right? There's still out of the molecules all those 1,000 molecules that I say will be good It's only one in a hundred that's going to be good, but because it's hundred after hundred better than random Well 10 out of a hundred doesn't start 10 up to a thousand doesn't sound good But the point is I don't need a thousand good molecules if I have one good hit. That's enough But a tenth of a hit is not enough So that's very much how you think I accept that 990 out of the test will be crapped because I get 10 good ones And then I go into the lab and when I found those 10 I will of course discard the other 990 That's all based on statistics That's a good question. So what was a pharmacophore in the first place? So given that definition what would you say so this is the trick right with pharmacophore We describe the molecule we describe the molecules that bind Which is both the curse and the blessing the blessing is that? Describing the entire receptor is difficult But if I know ten other small molecules that bind it's much easier to describe their average properties because they're small On the other hand the curse there is that if I don't know what's binding it's going to be virtually impossible to design a pharmacophore So you need either you need a very clear binding site where it's obvious where you're going to need something Hydrophobic hydrophilic hydrogen bond donor acceptor or you know five ten other molecules that bind So based on that how likely is that's an I know I keep asking questions But it's very can help based on that scenario how likely is it that you start out in the case where you have ten molecules at mind? well You to make a pharmacophore you needed a handful of molecules of bind right so I'm It's a chicken and egg problem here, but let's assume that we're starting out We want to make a pharmacophore. How likely is it that you know ten molecules that bind to your receptor any other suggestions? So think about what I said yesterday. What is the main problem in drug design? Sorry? Right, it's not a hard anybody can find a molecule that binds we can do it almost randomly And we even saw some of these databases right that you had a hundred forty six molecules binding It's trivial to find things that bind and I would Unfortunately, this is also where a lot of theory goes wrong both simulations free-energy docking and bioinformatics That theoreticians me included we think that it's so hard to predict that a molecule binds The hard part is making sure that it doesn't bind in the wrong place making sure that the admin talks works Lipinski's rule of five Yeah, it's great that it's binding to your receptor But if the molecule can never get to that receptor because it's not soluble in water Or that it's a peptide drug that's going to be digested. It doesn't really help us that it buys there, right? So that binding is not hard binding is trivial. So let's come back to these questions and discuss them But maybe we should structure it a little bit more Let's assume that you're hired in a research group at PsyLive app after your thesis project And this is actually not This is not just a theory of we have a bunch of research groups that work on very concrete targets saying cancer and You have in a tire scene kindness receptor or something that's this team is very interested in How would you go about it? Don't think that this professor is gonna tell you exactly what to do because this professor is this rehearse Specialization is in cancer research. They have no idea about drug design docking all this modern stuff That's why they're hiring you So how would you get started they have receptor X that they are interested in X marks is bought Talk to them about that with your colleague next person Well, you can have a team of three there spend three minutes talking about that and then we're gonna see how good you are at this Whether you're gonna get fired or We know I So I'll give you I'll give you one more slash clue slash instruction to think about The one thing that these research group has find out is that this receptor or whatever protein it is There's a mutant that is this mutation that is somehow causing this disease That's all you know so get that based on what we said where don't jump too far ahead Take a step back and think what the actual problem is and what you need to think the only thing You know is that there's a mutation in that protein and that is somehow causing Humor growth so we can also all the details to but again don't jump straight to the details So We discussed it a bit I hear some good ideas So you're now hired and we're sitting at this group meeting. It's a large likely large group They like to have 30 people in the room or something and they we need to brainstorm. What should we do? if So that and this is important, right? Don't even think about structure yet. The first thing you need to understand is biology This might sound obvious, but there is so much high throughput research now that you're only looking at four nucleotide bases on screens That yes, you can certainly find patterns there You can do whole genome analysis and realize all these patients have mutation in this particular gene But that's not going to get you anywhere unless you understand the biology. What is it doing? Why? So how could you understand the biology? It's in position 394 I just made that up Yeah, they grow tumors You you can certainly do that we don't unfortunately that's actually a course we don't have in the program sadly So there are certainly things like super resolution microscopy and everything The problem with that is that both that and growing these specific cells that typically require you to have cell lines and everything specific individual cells that express this And you can certainly do that. I bet the large you would do but there are things you can do already on the biofumatics level That also that you haven't you didn't explicitly study that in the biofumatics course, but it's related to these networks You can see what other proteins is this protein interacting with and there are databases for instance Have you heard about keg for instance? So there are database of pathways metabolism and not that you care about the sequences, right? but where in the metabolic pathways is this protein present and it might be if it's a If it's a tyrosine kinase we know exactly where it is, right? But in general you need to understand out of all the molecules in this class exactly what is this protein doing so Then let's assume that we are there and we do know the biology We do know the two three interaction partners this protein has and we you might even have a hunch what's happening in this process So what's the next step? Hmm on the on the other end if there were other drugs that worked on a very related protein We would likely not be doing the research, right so that we don't have any drugs We have we still don't remember the only thing we do know we know roughly where in the metabolic pathway this protein is Exactly and then this affecting the other partners. That's a bit fussy, right? But it is that's whatever is happening in this pathway. Are we too active or not active enough? the likelihood The likelihood that mutations create a completely new function. It's unlikely If theory can happen, but what virtually always happen is that? Sorry if the function of a protein was completely altered the organism the human would likely die But so the reason why it's only creating disease is that something is either too active or not active enough Let's say that it's too active It's up-regulating What can you say from that? Sometimes you'd plus this too For some reason something is severely up-regulated here Conversely, if it's severely down-regulated What would we need to do? It had to do with these classes of drugs we talked about, right? So what type of drug do you need in this case versus that case? Then we need an inhibitor to kill it Partial or full agonist, right? So we need to create for whatever reason your body is not activating this receptor well enough So you need to add some molecules that activates the processor, which the body does not do And that's why I actually Am I possibly I should possibly have emphasized that better yesterday? But that is why this concept because the concepts are going to be completely different whether you want to destroy a process or enhance it It's far easier in general. It's far easier to destroy things than to create things and most drugs we have tend to be inhibitors There are gradually more and more agonists And they're the simplest agonist you can imagine what type what would what would well a simple agonist At least what would you try to what strategy would you follow there this kind of two pads either? We need to Destroy the process or we need to enhance it So that's that's certainly one possibility is to get it finding something that works exactly the same way Now this is a bit of a trick question because there was something I did not mention in the slide yesterday agonists by agonists well Normal agonists they should bind in the same place. Are there other types of drugs you can imagine I mentioned it a few lectures ago You can find allosteric drugs, right? So maybe Maybe there is something that goes wrong inside the protein machinery So can I somehow add a drug that binds in a completely separate binding site? But that somehow rescues this functionality and the reason for that is that this process for instance Let's assume that is that is not gated by a ligand. It could be a voltage gated channel, right? I can't increase the voltage in the cell and the voltage that's not really a binding site But maybe I can create a drug that somehow helps the voltage activation and we when it's in terms of drugs We usually call those agonists to it will have the same effect But it doesn't necessarily mean that the process itself was activated by a drug and somewhere here Remember, let's get back to our receptor X. What what do you need to do in the team? Let's have some of the other groups answer questions to what would the entire team do That's whether it's an agonist with whether it's an inhibitor you would like to get or an agonist you would like to get All we know is that there is a mutation in the site That is related to regulation and it's our pet protein. So we're two weeks later Sitting in the group meeting and the brainstorming. We know more than two weeks ago, but To get there. What do we need? Yeah, but I said that we were unlucky. We don't know anything that binds to it yet We're gonna need the structure at some point We're fumbling in the dark here or we know that there is a mutation somewhere in a long sequence and we hardly have any idea what the protein is doing How can we get a structure? That's the right answer and why why why do you think that I would have been upset if you had not said homology models? Yes, then it would cost we can do an x-ray crystal, but it's gonna cost you one million dollars And yes, we can do that if we absolutely have to but if this protein had seventy percent residue sequence that your PI Would kill you if he realized that you could have done this in an afternoon rather than spend a million dollars Actually, it's not so much the million dollars that matters It might take two years to get that x-ray crystal and it's there is no question that you're sorry So if your homology model is not going to be as good, but the homology model we can have tomorrow afternoon And it's not strictly homology model, right? But in general any type of comparative modeling protea structure prediction as you learn to the bioinformatics Chris is starting to be Scareally good it used to be that they would laugh at you if you tried to do drug design with a model to models But that's all over the case But let's assume that your model is not that good then you need to do get a structure and How what would you do if you had to get a structure a real structure? Experimental one so the first thing is that you should not do it yourself Find a partner find a collaborator group who's really good at structure determination and this is what happens So that this big cancer research group at PsyLive lab would now say maybe call my colleague David drew or somebody here to see could they help you determine a structure? Because they're not you're not your own even something expert in determining a structure, and you don't want to wait four years So you say cry we am That's certainly one alternative. Are there other alternatives? x-ray, so there are advantages and disadvantages with both of these We're gonna talk a little bit about that this afternoon during the study visit What's the big and you might it might I like cream in memories, but it's certainly not the universal method What is the advanced you cry you? But that's not an advantage But right it's fast You don't we don't need as much protein if you can produce your protein We might be able to get some information about this in a month or two months The bad thing is exactly what you say this Resolution can be lousy and even even the structures that those of us to do cry em are ecstatic about there So it makes me good three angstrom resolution By x-ray standards that's a lousy structure At three angstrom resolution So what is if you're thinking about the Heidi Japan's resolution to get these things right the distance between a hydrogen and a carbon is Roughly one angstrom in a hydrogen bond the difference in possession between the oxygen and the nitrogen is 3.5 angstrom If you have three angstroms resolution, and that's the average resolution of the structure What is going to the quality of your binding site going to be? So in many cases the quality of a crye and structures in the ballpark of an homology model I So wish that was not true, but it is true in general The other problem is a crye and frequently can determine structures of small things so I Hope wish and pray that things are going to be different in ten years, and I think they likely will But for now there are very few if any cases where cry em has been used successfully in drug design You might be able to see roughly where a drug binds and we're developing these techniques I so hope that we can't change it, but for now you will likely need to go to x-ray Because if you have a good x-ray group, they're going to give you a 1.5 angstrom structure And this might not sound like a big deal But the point is at 1.5 angstrom you're going to see all the sides just all the atoms you see everything So with that binding sites we know we know everything we want about the binding site So bit boring, but sadly this is still the norm So what I would do I might variable call one crye in group and one x-ray group Or David drew who is doing both because then they can overexpress the protein You're going to need a factor of 10 more protein for x-ray But I would ask the chem. Let's try to crystallize it and see what happens because if we're lucky and we get good crystals Take them down to the synchrotron in Lund and then we might get an x-ray crystal in a few months On the other hand in six months when they are still struggling and they can't get the crystals to grow having a crye and structure is a whole lot better than nothing and Marta might cry a bit if I told that that's well, no she knows about it Okay, now you have a structure and we're six months later into the product now because that took a while But no but but this is important right that drug design does not happen overnight These are very large campaigns in groups and in some case It can take more than the length of a PhD to target a new receptor because if this is successful We're going to be able to cure new forms of cancer Have a structure. What do you do? Sorry? Sorry Well, you can't really patent the structure. Why can't you patent the structure? That's actually a very good question Why can't you patent the structure so that to be able to patent something? It has to have technical effect and it has to be an invention And it should not be an obvious combination of two results so the fact that an obvious is a bit sliding I Actually worked at the patent office when I was a student over summer. Let's let's come up with a really stupid invention When you're doing the dishes and you need a brush, right? But why should I hold that brush? Can we immense on like take a glove and then put lots of brushes on the glove so that you can just put on the Glove and and then do the dishes with the glove. That sounds like a cool invention Do you think I could patent that that's my friends is a class in the patent system There are like 500 patents in that class That just because things are not sold on the market doesn't mean that somebody didn't patent them 50 years ago a patent Is not guaranteed that it's a good or a smart product or that anybody will want to buy your product See in practice the thresholds for patenting things is fairly low What we do say is that you're not allowed to patent the result when you know I'm not allowed to patent the sound because it's not my invention. It's something that exists in nature Just that I discovered something doesn't mean that I can patent it But there are a couple of examples. Have you ever heard about myriad genetics? Have you heard about this? Did you hear about this is a bioinformatics course? PRCA So what is that? It's a horrible breast cancer gene so that if you have this gene. It's only a matter of time You will develop breast cancer And that's why there have been a number of Celebrities and everything that you do preventive mastectomy you remove the breasts You might even remove the womb by the time you've had kids in the 30s or so because to prevent you from getting breast cancer So, how do we how do we know that you have this gene? Adrenaline myriad genetics patented the gene And then they sold the test at the cost of only $50,000 They have a patent on the gene so you're not allowed to develop a competing test This has made its round through the courts in the US and I think what eventually happened This was eventually turned down in the Supreme Court that they cannot patent the gene But what people try to do you patent the techniques by which you sequence this particular gene Or you try to patent all the details about this specific test It's the same strategy to create a bomb carpet make it very difficult for some of them But formally you can't patent the gene. You can't patent the sequence. You can't patent the protein If it's naturally occurring Because you have the right to your DNA, even if you share my sequence. I can't patent I can't patent sequences your DNA In theory at least So you can't patent the protein, sorry And we're still sitting here with our structure and your PIs don't get a bit irritated because you just spent a million dollars on the X-ray structure And we just have a structure. We're not closer to any drug That's that's a good idea and ideally what we likely did when we sequenced this protein, right that You probably took both the wild type and the mutant and tried to determine the structure of both of them at the same time So let's assume we did that But it's actually it's a very good study. You said if you didn't do that you could if you have the structure of the wild type, right? Getting now making the homology model and predicting the effect of that single site mutation That's something that the computer should be really good at because you have like 99 percent sequence identity You could even do a simulation and try to understand what this particular mutant does Let's assume that it changes something in the pocket here. So we have the binding pockets, whatever Let's say that this is the wild type and Then we have the mutant that it will look different It's a valiness that a felony. It's a different shape in the pocket So now I might have jumped a little bit here But but assuming that this was some protein that the activity was too low, right? Can you guess what might have happened here? Before we start talking Yes, either it gained affinity to other molecules or the molecule that normally would bind here is now too large to fit Or something so that for whatever reason then nor the molecule that would normally activate this receptor can't really bind there anymore Because we the binding site has changed a bit. Yes, or or the opposite The other alternative if it was upregulated it could of course have been that because this has changed There would now be too many smaller molecules binding or something but something has changed in the binding properties here Again, your first bet should be even if you don't know what the binding site is when you have a structure You know where on the molecule this mutation happened And the first try your first strategy should of course to be to look at that region the likelihood that there is something on the Other side of the protein that's changed is nil and then you mentioned some strategy. What say is that you wanted to do here? docking yes So at this point you start doing what you call and docking there is another abbreviation for that that I actually I told you about it But I didn't tell you the abbreviation You will occasionally see that what does that stand for I said it yesterday I didn't I didn't spell I didn't say that I didn't tell you about the abbreviation. So what does HTS stand for? High throughput screening and the V stands for virtual So because traditionally you might do this in the lab, right? But the virtual means that we do it in the computer and the virtual high throughput screening We do by docking and then we do exactly the thing that we told you that start from a million molecules If there's one database I had to pick I would pick zinc Zinc is not commercial if you run a large pharmaceutical company and let's say that you are If a large pharmaceutical company that's specializing on cancer They likely have their own internal molecular databases with a million drugs that they know that they tend to be important for these receptors Do you think they will share these databases with researchers? So that do you think they will make them public? Because this database is stuff that you can patent they likely have a patent on every single molecule in this database So this is something they want to hold close to the chest on the other hand pharmaceutical companies They only have limited manpower inside the company and everything so they would likely like to work together with researchers So at this point, what can you do if you're in this research team? You can contact them right and say that you would like to collaborate with them And they might at this point you might do some get some sort of agreement The research groups would probably make very clear that they own the rights to the drug and everything But maybe the pharmaceutical company can help you screen Because in the first round maybe you didn't find anything in this open-free database, but they can also test their drugs and If you're then going to publish that you're not going to publish the structure of those drugs So you can say that you tested this on all of zinc and you found these ten hits And then you also tested it on or you might not even include that in the paper So or you might say that you had this other database and there were four compounds named a to f But you're not going to tell for now what those compounds are because you're going to try to patent it Why can't you first publish and then patent? You're only allowed to get a patent for things that have not yet are not yet known So you need to submit the patent application before you submit the publication. Good. Let's say that you found five heads here And they have some reasonable binding affinity what do you do? Yes. Well This is something you want to do before you had to in vitro screening maybe or actually no, you're probably right, but that You might it makes sense that you want to test them But but I said that when we found five let's say that we found the docking selected one thousand when I tested this in The lab I found five good ones or five ones that had some sort of effect on the receptor What do we need to do by these with these five well long before simulation in General if I don't tell you anything how good you think those are going to be Exactly right there like in general. They're hardly going to show. It's just 95 it's just what slightly outside of the standard deviation So there is some sort of significance here, but it's not very strong, but it's the best you have Alter them. How do you how would you alter them and there are we mentioned a couple of different strategies this right? But this is all optimization page. So now we had some hits the hit just mean that we found something that works and And now we're in this place where you call lead optimizations The hit the hit is now something that we want to let us literally lead a clue that we want to iteratively improve and Try to get Can we get that to be better than just so slightly significant to be something that is a really good binder? How would you do this? Well, the first thing that's again something I did what you would likely contact some organic chemists say upstairs here People are really good at designing molecules and it might for instance the solubility of your molecule might be too low Can we improve the solubility you ask the organic chemist to help you with that? And then you get some new molecules and then you try them again And you might even use some computers even some you might more or less randomly try to add or remove small groups on the molecules And this of course experience counts There's even an entire field called it for the called medicinal chemistry And that's really when you try to alter the chemistry of small molecules. This might take you a year Gradually improving that but at the end of this year, let's see if you have some sort of plot if this was wild type You might have started out here. That's something that was Just a slightly significant and then after a year we have a gigantic very strong signal. It's very significant It's a good binder and everything What do you do here before you go to clinical trials, which is called pre-clinical, right? So you do animal tests But at this point the research is almost over At this point things starts to become so until the states it's cheap and now it starts to be quite expensive Because you don't want to run an animal facility. Trust me So at this point you're gonna need other teams involved Actually, the the the animal test will likely be done in this large research group, but with somebody who's an expert on testing this in mice What do you use mice? It's cheap fast, right? And we don't need that kind of payment I think I already mentioned this joke to you that If you're a mouse and get cancer, we have a very good treatment for you And at some point there might be some more iterations here You might want to try to remove some toxicity you might try to but the point you can't change things too much Because that you actually starts to take steps back here. You're gonna need to redo everything you did So it's not strictly a one-way street So at the at the animal testing you might still want to try to improve or you might have ten candidate molecules and Based on these ten candidate molecules slight variations. You might end up picking two or maybe three possibly even just one What do you do with that one molecule? So the first thing you do is actually wait for what you said here's the patterns So now you that you've shown that this molecule works. It has a technical effect on something now You pat them before you submit the publication The second thing you do you would like to create a small startup Why would you create a small startup? And it's well, it's also if it starts to get very expensive now, right? The likelihood that you will be able to afford the next steps in academia is fairly low So you create a small startup and then you approach a couple of people Say venture capitalists and everything say you would need to take this through clinical trials here And this is going to cost you a lot of money and then they invest in this. This is super high risk investments It's another famous quote by Peter Lynch who is the famous very famous stock investor and everything he used to The one type of company he never invested in was biotechnology or medical startups because they say that There were 100 PhDs 99 microscopes because there's one CEO and it doesn't have a microscope and zero profits Because the point is you're not you you don't have any income whatsoever in the company. You have nothing to sell But you're gonna need to spend ten million dollars to take this through testing So they so why would people be stupid enough to invest in a company like this? If it works, it's a hundred billion, right? So that it's super high risk investment that they even know that on average They're gonna lose their money, but one company out of ten is gonna work so well that they get it back What typically happens today is that it used to be the fact that the big pharmaceutical companies bought them with your patent or something here What typically happens today is that they are too conservative and they know that there are so many things that fail So on the very first stage, you typically have researchers maybe some funding from an innovation agency Maybe is some external capitalist. You typically take things through phase one yourself or in the small company What was phase one? Healthy people you almost it doesn't kill you. Hopefully and Then somewhere in during phase two or at least when you start getting the first phase two results That might be where a larger company buys you So what was phase two? Patients and you need to show that it actually has an effect The other difference when you're testing things in phase one, you might have a few dozen Patients or so so they still cost money because you're gonna need The regulatory paperwork is quite intense. You're gonna need doctors on board and everything you need lots of permits You need the money you're paying to the say the students or something is a very small cost But getting all the medical staff involved starts to cost money, but it's still doable Why is phase two so much more expensive than phase one? So well, how do you convince the doctors or it's so much the regulatory agencies? What is that they want? They want statistics, right? So in phase two, you're gonna probably gonna need at least a factor of 10 more subjects Maybe a factor of 20 more subjects and that also means that it's gonna cost you 20 times more money because now you're gonna need You're gonna need to test this in different parts of the world You're gonna need to say that if you take it does this drug work in men in women, etc young old people Again, there's lots of statistical requirements and now This starts cost starts to go up and you're 10 million dollar You hadn't started funding is depleted in your company and you can't take it any further And at this point you frequently have a medium-sized company that comes in and say that we will buy your entire company for 50 million dollars Which is a pretty good turnover right that you the investment might have been five million dollars and now they're buying you for 50 million dollars But they won't use researchers in so that they basically they buy the entire company in the stuff This song will act we hire you actually it's just as much the staff as the drug you're buying But they buy all your patents and everything this company would then take this to phase three and phase three What what and you need to show that is better, right? You can imagine the statistical requirements are now even higher. So it's another factor 10 more expensive And what frequently happens in phase three is that this medium-sized fish gets bought And even larger fish and then you have fighter Azra Zeneca and they might now spend a billion dollars in your company because this looks promising that of course the second The phase three is ready. The second this drug is approved, right? It's too late Because then you have they can take it to market. So they They frequently look at the statistics and look everything here looks so good So we think it's very likely that this company will succeed. So we'll buy it now while they're cheap So it's increasingly happened that more and more development happens in the quite small companies and in particular in academia And that's why this is not a fake example. You will Quite a few of you might where we'll start working in case like this One problem here is that one problem here is actually that Throughout all this testing From a pharma big pharma is frequently criticized. So one thing that they're frequently criticized for is that there are more drugs that are developed for Male diseases or that there are better indications for men so for anything that's related to congestive heart failure or something We frequently tested better on men than on women Why? No, they're well, there's someone's related to this and in particularly the early stages there Well, we can certainly tested on female mice, too There are both female and male mice the last time I checked But there is one problem there is one problem in this early testing that we haven't solved Because that as we mentioned where there are dangers for drugs And there is something that don't happen to men that can happen to women pregnancy And the problem is in those rare cases when something goes wrong, right? Actually, it's not we're not talking about the healthy side to the patients and everything If something goes wrong, it might cause genetic defects or something in the next generation And of course if we give a man a drug and they fall a bit ill, right? We can we can stop administering the drug So that what frequently happens is that you first it's an order of magnitude more difficult to find side effects related to pregnancy Particularly for the baby So that's what frequently happens. We first get the drug approved for the men and when we see it out to two or three years that they don't die Then you might also allow number because and we're talking about very small statistical effects, right? And then when we've had the drug on the market two three years, we might also show that We might also dare to start making larger and larger scale trials and pregnant women And it's not until you have tested it on pregnant women that you're comfortable giving it a pregnant women And if you think that this is stupid think about Thali to me, right? So that we've had some pretty scary things in the history of science where there are horrible birth defects and everything caused by drugs But I think this is roughly how it goes It happens all the time and there are tons of new drugs developed But now we spoke about humans The problem here is that this process takes 10 years and a lot of development now is not necessarily this type of drugs Because if you're a company as you probably saw here the risks are pretty high, right? So the traditional way would be to focus on medicine and there are still a lot of companies doing that But a lot of development today is what we call biotech instead and Companies tend to prefer biotech in particular small companies for a few reasons So can you imagine anything you would do like to do in biotech? related to Understanding binding or something I spoke a little bit about it at the break yesterday with the future for instance What if can you get bacteria to produce ethanol or something or terpenes or something? So we can we produce biofuels in bacteria then you might need to alter the bacteria bit It's not we're not literally developing drugs against any mutation, right? But we might want to create mutations to get the bacteria to produce something different Something else you might want to Producing a better insecticide or an insecticide that kills the Some sort of insect, but it doesn't hurt bees. It's actually a significant problem now that we hit we kill bees with insecticides One advantage here is that you can imagine again pregnant women on one side versus insects or cockroaches on the other The requirements when it comes to cockroaches are not quite as high So that if you have a new insecticide you might where we'll be able to go from lab to the market in one year Which beats ten years So in general biotech is nice because or you can imagine sort any type of better product to improve an industrial process or something The advantage is that you cut the development time in ten There are no clinical trials with ten thousand patients involved There are no regular hardly any regulatory agents You would probably need to get the insecticide approved But again the requirements are orders of magnitudes lower in the patient so that a lot of small companies today tend to focus on biotech rather than medicine That's roughly what we had. Is there anything else you want to say about that or should we head on to free end this? They can be highly profitable It depends there are certainly bad ones too, right? Many biotech companies even might even focus on developing things that are Important for the pharmaceutical companies, but it's not the directly the drug one example could be new delivery mechanisms So can you create some sort of small vesicles that help you deliver the drug to the cancer cell? So we're not developing the cancer drug, but we're developing some sort of vehicle that you can use for hundreds of different drugs to develop them So they're very profitable. If you look in the Bay Area, for instance that Biotech is kind of the new digitalization area There's a huge amount of things going right to digitalization of life sciences And it doesn't mean that we turn everything into bioinformatics But as I mentioned yesterday, we're increasingly having robots steered by Python scripts and everything which has things on a very large scale So it's a it's a it's a it's a large industry. That's growing and it's very profitable. I Would even say it's far more profitable than the computer industry because the problem with the computer industry is that things There is a war that the price war drive also drives the profits towards zero So it's very hard to make money in computing because you're in better You're competing with Amazon if you're competing with Amazon Google and Facebook. It's hard to make money Biotech is more of a Klondike with thousands of small companies than many There are certainly quite a few that go bankrupt, but many of them make a lot of money There was a large scandal in Silicon Valley a few years ago I should know the name here And now it escapes me, but it's basically a new company that proposed that they could develop a broad spectrum of blood tests and using new fancy biotech that they could do this with one tenth to one hundred of the amount of blood They do very efficient genetic testing and everything And it turned out to not it wasn't quite a scam But these methods never really worked and eventually the food and drug administration cracked down on this and Put them out of business But for a long time this was one of the hottest companies in Silicon Valley with a valuation of fifty billion dollars or something And now they're their background Because then the promise that it might the problem with the valuation of these companies is so much It's more based on future promise rather than current performance That if this might be the next cool thing that's going to put all other companies worldwide out of business and you have patents on all the key technologies The market goes crazy On the other hand, it's easy for us to say that on the other 15 years ago There was a small stupid search company that was started by a couple of students And I remember because I was a postdoc at the time and I thought people were completely insane to buy the stock at the IPO And this was Google And since then the stock valuation has probably gone up by a factor of hundred or a thousand Like I said it at the time it sounded like it was very overvalued. So don't take stock advice from me It's hard to make predictions in particular about the future free energy Yes, so oh, yes, sorry handouts for today Give you one two three there So why now that we've just talked about all these things and say that drug design kind of works Why would we need to head back in the computers and start doing some of these things in computers instead? Well, the reason is that drug design works, but it's it's exactly the things we mentioned. It's slow It's inefficient and it's expensive And there are also some things that are difficult to do simply again our this small lab that you were working in you might not have the equipment to Do all these tests you also just Organization-wise it's very difficult to have these large teams go doing drug design and the advantage of computers for all their other Faults and they do have numerous ones computers get roughly twice as fast every 18 months There is no other experimental technique that can compete with that possibly sequencing, but that's it So as limited as computers have been traditionally things grow exponentially And if you could now take many of these all these tests all the complicated things that we had to outsource And if you could run them overnight in a computer instead, that's very attractive So that then I'm not just saying this theoretically so that the company For instance these or research That I mentioned David Shaw they also own a very high stake in another company called Schrödinger and Schrödinger is a company that developed These computational chemistry tools There are a number of famous investors including both Bill Gates and Paul Allen That have invested a few tens of millions of dollars in each of these technologies So that these companies are breaking even they're even making profit, but they're not yet making gigantic profits But I think this is gradually where we're heading So the idea is to rather than going top-down that in a few cases, can we go bottom-up and use simulations models and everything to derive information about free-enders and An obvious way to do the binding free-enders if we could predict how well these drugs that we talked about this morning How well they're going to bind? But there are a bunch of other things that maybe we would like to get the free-end use of those transition states Right to explain how fast or slow things happen Maybe we want to calculate how soluble different molecules are in water This is something that's pretty interesting to the pharmaceutical companies If I could do this in one minute if you can take those one million molecules in the database rather than testing them I could calculate the solubility and Then we don't need to spend a date per molecule testing how soluble it is then we know what molecules are soluble enough So that we know we only need to test those The binding I already mentioned and there might be some sort of reaction coordinate That is what happens as a molecule is moving or as I'm pulling in a molecule or something I will show that with a few slides later on this can be easier I will tell a little bit about something called free energy cycles, which is related to mutations and Then towards them we're going to speak a little bit about calculations in simulations and our research so you already know the background to their free energy that The free energy determines the relative likelihood of states And this states could be the fraction of bound versus the fraction of unbound. I Already mentioned when we talk about membrane proteins that you will frequently invert this that we can talk We could measure the fraction of a membrane protein that was inserted versus not inserted And if I do that I can extract the Delta G between them just by taking RT-LN The difference the quotient of the probability And that's why you see this RT-LN case everywhere the case the equilibrium constant is literally going to describe the relative probabilities of being on the left versus the right-hand side and We also know by now that all chemical reactions will follow the path that will go down free energy wise so that if we can calculate free energies You can calculate what happens We might not be able to say exactly how fast it happens that will also require us to calculate the barriers But if we the point if we know free engines we know everything about a system and that's why it's somewhat like a holy grail And you know this that we've looked at this is statistical mechanics You looked at this is statistical sorry in MD simulations, and if you haven't done it yet in the lab I never remember the order of the labs and how they were synchronized But you're going to be looking at this a little bit both with docking and simulations in the last few labs here The problem though is that while it's easier to get the enthalpy here That's something I can calculate from one snapshot to get the free energy I also need the entropy and the only way you can get the entropy is To run a small simulation and let the system sample this so that I effectively get the probabilities The probability of being a particular state and here too that said this p delta v term We always ignore it because it's not important for whatever we do and unfortunately The delta s here is well there is no simple way to directly get the delta s from a simulation on the other hand If you think about it the opposite way if I if I run a very long simulation And I calculate what fraction of the molecules are in the water versus what fraction of the molecules are in oil I effectively get the probabilities right so I can certainly get this the back way from the probabilities So if you bear with me a little while here I'll come back to that in a second and say oh yeah, so a simple way is that you take the solute and Imagine that we have it in vacuum or air air is virtually the same as vacuum And then we take this small molecule and put it in water and then we would like to calculate how much did this change the free end Virtually all these parameters we had in force fields have actually been obtained the opposite way that we know for instance How much what is the free energy of solvating say benzene in water? That's a side chain of phenylalanine and from that we can parameterize phenylalanine And that we covered already the first two lectures that this is super important protein folding So they're one of the reasons why we can simulate proteins fairly accurately Is that we have cheated a bit and use all the experimental properties to parameterize our simulations? We don't try to get this from quantum mechanics But you can take if we now know the partition coefficient of that small side chain You could also calculate how different is it to put it in water versus the membrane? We can calculate that reasonably accurately or if this is the side chain We can calculate how likely is that the side chain is facing the solvent here Or that it would be on the inside of the protein and again here. You're just really talking about probabilities If you simulate this long enough I could see what fraction of if we spend it one percent of the time It's going to be here 99 percent of the time is here that I can calculate how accurate it is We even had I even had a student one of my first students years ago That started out to look at exactly these insertion free energies when we had for instance arginines at different locations in the helices To understand all these results that we got from the strange membrane insertion and it works quite well You get curves based on this that describe if you have a glutamic acid or arginine How much does it cost to put this in different positions in a helix? So the point is not that it's exactly accurate But we can use it to understand what happened in a particular how much water they are pulling in the membrane and everything But this is still on the fun to understand by physics point of view What we would like to get is get some sort of free energy of binding If you have that small molecule bound to your tyrosine kinase receptor What is the binding and kilojoules per mole or you can measure it in concentration, too? And there is actually if there's actually a fairly easy way to do that the free energy of binding is the amount of work that We gain basically I Can take that molecule and if I slowly pull that out the work I need to do to remove this Well once I am in the state where it's fully removed. This would be reversible, right? The way I got there is not important The free energy difference between Tuesday the free energy is a state variable So if I have a way to smoothly remove this molecule From the protein and if I can measure how much work or energy I have to add when I do this That should be the free energy of the process This is a bit fussy because we it might not be obvious how it and but I would are either one way Is if I just put put a small spring here, and then I start pulling in the spring I know exactly how much force I'm adding because I literally pulling in the spring myself I let the computer pull and Force multiplied by distance is energy, right? And then I know how much energy I had to add To pull this out There is one problem here if I do this very fast what's going to happen. I generate heat right friction So if I take a molecule and move it very fast through water I'm going to be generating a lot of heat in the water and The point is that now I'm not at exactly the same state So this works if I pull this infinitely slow I need to pull it so at least of course I can't put infinitely slow, right? But I'm going to pull it so slow that I'm in equilibrium all the way So that I'm not really heating up the water to any significant extent And if I could do this very slowly I could you could imagine doing it the other way Once I'm here, and I gradually pull it back if I can go smoothly between two states I should be able to measure how much force I added and Let's see I have an example here. Yes, if I had to add a lot of force here, right? Sorry, if it's a strong binder I had to pull with a lot of force and the integral under that curve would then correspond to the energy While the red curve here would be a much weak binder this works it's not what we normally do in simulations, but for a few cases we absolutely need this and If you think that this is artificial in a simulation It's not you can do this in an experiment too with an atomic force microscope and there are some really cool examples with EPCR proteins in particular So that an atomic force microscope is really just a tip it is a very narrow tip and This tip is so narrow that that the very edge of the tipper you just have one or a handful of atoms with and Then on the backside of the tipper you have a mirror and then you shine a laser on this mirror And then you're detecting if this cantilever moves because the sample here moves up by just a few angstroms You're gonna get small small small deviations in the mirror here, right? And based on that you're gonna hit different parts on the photodetector These are surprisingly cheap and simple markers. It's not really a microscope that you're seeing things, but we can We can get this tip just scan over the small surface And the type of experiments that people have been doing there You have for instance, you can take bacteria rhodopsin and pull it out of a membrane It's very cool to take some of these proteins that are involved versus in muscles These proteins that were elastic that we talked about we can literally get them to attach to this tip And then I'm measuring how much force am I applying when I'm pulling this apart. I Think this is even an example. No, sorry. This is not a movie So this is one small part they take a small molecule bound to want these proteins and then we attach the small spring to it and then we gradually pull this out and Then we did this in simulations. You have very noisy forces here And then they try to pull with different rates and everything and then they argued So that when it's very high rate here The force ends up being very high because you're generating lots of friction and Then they're arguing at some point that starts going down and then they're Extrapolating here that if we could pull infinitely slow, we would be have a force somewhere down here it kind of works and It's fun because we can actually gain information about for instance How much energy would it really require to pull something out of a membrane? Or if you have this chain the nascent chain in a ribosome The ribosome is kind of pulling on this chain right because it's the chain is being pulled through the ribosome How hard is it pulling on the chain? And we can measure that both in the lab and in simulations and get roughly similar results depends on what you mean by accurate We're not interested in getting the force here with three decimals, right? If it's a large system and you're talking about understanding you want to know are we talking about piconewtons or Nanonewtons or new tons and it's going to be closer to piconewtons The other thing that you see up here. Is this accurate? No, it's very noisy, right? But the free energy doesn't depend on the force there the free energy here is going to be the integral of this and You might be aware that it but anytime you integrate functions the integration has an averaging effect So when you if you look at these there are three curves here, right? So the solid the dashed and the dotted and They all fluctuate around the same values I would bet that the integral of those three curves while the curves vary the integral of the three are likely within 1% of each other So that by the time you integrate them the differences are going to be smaller than you think But that sounds bad. There are errors of 1% or something. How will that compare to an experiment? So that experiments are not as good as you think what do you think the error is in an experiment can easily be 10% If you're measuring one by I'm going to come back to binding energies in a second If you're measuring binding energies of 1 million molecules, how many times do you think you do it each experiment? Well, unless you do it multiple times, you're not even going to have standard errors, right? Let's be generous. Let's say that you made it three times And then you might get the binding to be five kilo calories per mole plus minus three kilo calories per mole You can have lousy experimental results. So in many cases the experimental error is larger than the computational error So these this is important if there are large changes in the molecule or that might be a large receptor Then I really want to happen what happens when this entire receptor opens And then I would need to take the entire protein and force the protein to undergo this transition My ion channels is yet another example I know that the ion channels will have to open up and can I force the ion channel to open up to understand the free energy difference So maybe I have an close state here and then a barrier and then an open state And this would correspond to the radius of my channel And I can do something like that by forcing pushing the channel open and measuring how much force I need to do And then okay, so there's one stable state here. There is a transition state and there's another stable state here And when it comes to my ion channels what I'm interested in is What is this Delta G1 and Same thing. What is that Delta G2? And what I care about here is not whether it's 5.39 or 5.38. What I care about is this 2 or 20 Because that's going to influence how quickly the reaction happens, right? So frequently we sometimes talk about qualitative results, but qualitative results are usually more than fine even for designing a drug So the concept here Is universal to get the free energy between the two states They are state variables, so the free energy can only depend on the state and To measure that or to calculate it I need to measure how much energy do I need to add or remove when I force the system to move between these two states And what I hinted But this will only work if I stay in equilibrium all the time I do this Because otherwise I keep adding heat friction or something so I need to do in general. I need to do this very slow This is actually a universal result. So anything I have For instance these small molecules I talk about this FKBP protein yesterday If I want to calculate how much energy do we gain by binding that particular molecule here There are two states I have here. I have on the one state I have the protein without anything bound and the small ligand in water and In the other state I have the protein with the ligand bound. This particular ligand is called FK501 And it's a pretty boring name for a protein because the protein is named after the ligand So this is FK binding protein So the only thing we need to do to be able to calculate is we need to find a way to go between these two states If I can calculate how much energy I need to add or work I need to do to move between that state and that state or between that state and that state I can calculate the free end at a point here that the path does not matter if the path mattered they would not be state variables and Here is where we can be smart There's something that says on the very last line here Simulations do not have to obey the laws of physics That sounds horrible And it's not quite strictly true. Somehow I Need to obey the gen- the simulations need to be specified that From the laws of physics, but they don't I don't necessarily need to restrict simulations of things that can't happen Let's let's see if I can come up with an analogy here. Well, again, let's use the example of Me we want to measure how expensive it is for me to go up once there And the obvious way is to calculate my weight Which I have no idea what it is probably hundred kilos out and raise that energy by let's see five meters or something So I need to walk out and I need to go hundred meters in that direction Take the stairs and go back here, and then we would be hundred kilos five meters total up That would be a way of measuring that free entity or I could Transport well calculate the quantum chemical process if I tunnel through the roof here What result will you get the same one, right? So while it's obvious and in this particular case, let's assume that it was easier for us to calculate the tunneling It's not gonna be The point is I am allowed if I only want to know the difference I don't need to take the same path that nature would take if for some reason For nature, it's of course easier to take the stair path, right? But what if it was easier in the computer to calculate the tunneling? So what I did before when I'm gradually pulling a molecule out I'm doing exactly the thing that nature would do I would unbind the entire molecule But what if you have two gigantic proteins? So here's protein one with 1500 residues, and then you have another gigantic protein with 1500 residues here and Here we have that small mutant that you had Trying to pull these entire proteins apart. It would lead to gigantic free energy differences, right? But the only thing that we were interested in did that mutation help or hurt the binding? But let's assume that you could calculate this. Let's assume we can't pull them apart It's gonna be slow in a computer, but I can't do it and the result is 1593 Kilojoules per mole for the wild type The binding energy, sorry the minus, it's binding. That doesn't tell you anything So the only thing that helps is if we can do this both for the wild type and the mutant, right? Let's do it for the mutant too. It's minus 1596 Kilojoules per mole. Wild type Alanine 294 V. So what would you say based on this? The mutant, sorry, it's negative here. So it's going down is good So this appears to be the mutant being the mutant would enhance the binding Because negative binding, binding energies are negative when they are good. Oh, sure When yes, you're right. And when I when I pull them apart, that's why I added the minus I when I pull them apart, I would need to pay 1593 in one case, right? But when I report that I report it as the binding energy so that I need to change the sign there Well, it might appear that way, but here's the problem. You don't get 1593 Do you really think you have four digits of accuracy in this result? So if there's gigantic things when you're pulling them apart, the real way of the reporting the first result would be 1593 say plus minus say 10 Kilojoules per mole And the other one would be 1596 say plus minus 9 So the real result we have here, I'm not gonna do that I'll come back to the way we combine this but the the difference here then is roughly 3 Plus minus, let's say the ballpark of 10 So you don't even know what the sign is The problem is that you're taking your this is a very small difference between two gigantic numbers, right? And while compared to 1593, this is a great standard deviation It's only one part in one part in the hundred. That's wrong But because the difference is so small now the error is but the ballpark of three times larger This would actually be probably be 13 or so because you're gonna add them with the square root So this doesn't work at all and this is frequently what you end up in when you try to come you can never you comparing large numbers doesn't work So what we need to do we would like to find a way that if you have a large molecule bound If I have two molecules that are quite similar, what is the difference between these two molecules? I would tell you what is the difference of that alanine 394 v mutants? You would like to test only that without completely pulling them apart Before we do that. Let's look at one molecule It makes it slightly easier So normally one state you had is that you had the protein and the molecule and they are not interacting and on the one hand We have the protein and the molecule and they are interacting But to go between those two states I would need to pull the molecule to or from the protein and if this is a large molecule That's where things go wrong So let's draw a couple of states up here up here We have the protein and the ligand bound and that's one of the states that I want to be in Let's just for a second assume that I take this ligand and I gradually turn it into a ghost I Gradually disappear all the atoms. I first it's 99% and then I gradually turn down all the atoms to see I just turn off all the atom interactions You can't do that. Well, I can in a computer and of course when I'm here the ligand is present to 50% That's a completely unrealistic and unphysical state But I am allowed to do it in a computer and when I'm down here I have the protein and then I have a molecule that doesn't really exist. It's just a dummy. That's not interacting with anything Sorry, it's basically these are remember these are just coordinates in a computer But when I've turned off those coordinates completely, they they don't exist They are just points in space without any interactions. Then I just have the protein here This would then be the case when I just have the ligand in water So think of the having taking the ligand and pulling that away from the protein You can certainly simulate the ligand in water in a simulation, right? But I if I have that ligand in water, I could gradually turn off the ligand in water to just gradually remove all those interactions That is also well-defined states. So then I just have water with nothing in it So each of these corners is well-defined. This means that I just have here. I have the protein and the ligand In water here. I just have the ligand in water here. I just have water and here. I just have the protein and What I really want to what I really want to find out is that how much does it cost when I move the ligand to the protein? But that's hard because then I would need to pull it away But let's say that if you accepted for a second that in theory I could go between all these states in a computer If I start out here and then I go there and then I go there and then I go there and then I go there What is the free energy difference? Zero because I'm back where I started and free energy is a state variable So the smart thing here is that the difference between these two states Must be the same as the difference between those two states the sum of the entire Circle here must be zero and again this dummy I'm not really changing anything here because the protein would here in the water. Sorry. I didn't draw that So here I would have the protein on the water too. I'm not changing the protein on that side so that Here I just have the protein of the dummy and there I just have the dummy in the water I haven't changed anything there so that with a bit of argument you can argue that that's going to be zero So while this is what I would like to calculate this is hard to calculate in a computer But these two pads are kind of simple to calculate in a computer. They're not physical. They're horrible things I'm gradually turning things off, but I can't do that in a computer. I can't stop halfway Because when I'm halfway in and I'm in a completely unphysical state But when I've gotten to the other side that is a well-defined state And so this corresponds to calculating me tunneling through the roof here. I can't stop halfway But once I am at the other state and if I can't calculate this efficiently the result will be valid and Now I'm not calculating the entire protein I'll show you one more slide before we take a break because every life this sounds a bit confusing now, right? We normally don't do this for one leg and but if we could This would create the absolute difference in free energy of binding But the problem if you think about large molecules here, that would still be 1500. It's still a very large number And then I could calculate for one more ligand and I will get another very large number and we're not so interested in this 1593 I'm interested in what is the difference from my small either a difference in an amino acid or if you have an entire series of different drugs What is the difference? So we're virtually always interested in difference in free and just not the absolute free and yes But actually this binding energy that per se was a difference in free energy, right? So what you typically want if I want to see I want to see differences of differences I will repeat this after the break to but I will at least introduce the concept to you now This is not as complicated as it sounds as it looks My two molecules the cost of pulling them apart was a difference in free energy, right? That's a delta G But what the real question is the question we wanted to ask if you swap that alanine for a valine How did that change the binding energy and that's a difference of a difference? So the delta delta G course once when I make this mutation Do I improve or deteriorate the binding and the smart thing here? I can do is That I don't really need to pull the entire molecules apart So if here's a protein and then another receptor protein Let's just Gradually change the alanine to the valine here and then I take it when they are so first I do this in one case when they are bound I gradually change an alanine to a valine when they are bound That's a tiny difference. It's just like 10 atoms. I'm changing out of an interface where we have 500 residues the computer is going to be really good at doing that and then I Remove this I protein and I do exactly the same calculation, but just with this say the left-hand protein. I Gradually change my alanine to a valine. It's also a very small difference. I'm calculating there So let's say one of them might be say three Plus minus. Let's say that the error here was again one in a hundred. So three Plus minus point over three kilojoules per mole And in the other case, I say that it's going to be six plus minus point oh Oh Six kilojoules per mole But here, too, you have one state when both of them are in wild type bound I have one state where I have the mutant in the bound complex. I Can think of removing that and removing that here, too This is a cycle if I take the entire cycle the difference must be zero And here's the beauty What I was interested in how much does it cost to bind the wild type Compared to how much does it cost to bind the mutant? Here's the bind that process is binding the mutant that process is binding the wild type, right? Both of them were plus minus fifteen hundred the difference here Will correspond to the difference here But those two process are trivial to calculate So let's do the math here again. The difference between these would be say roughly three plus minus This would be maybe something plus minus point. Oh, oh eight. Do you see what I get? The error here is like one hundred times smaller So now I can definitely say that one of them was better. The error is just point one kilojoules per mole I realize it's a bit confusing Think about this during the break take 30 minutes, and then I will go through it once more in detail Yes Exactly that you need to be a little sorry I was a little bit sloppy here You need to be careful with the signs here by because the free energy in one direction here It's going to be minus the free energy in the other direction So I didn't bother about the sign we can do that after the break if you want to but the point is yes So the three here is one of them and the six here is one of them these two would be the 1300 and 1300 plus three So the difference between them is the same but both of these are going to be gigantic free energies that are difficult to calculate Both of these are going to be small free energies that are easy to calculate And now of course these are fake transitions. They could not happen in nature I can't swap an alanine to a valine inside a protein or inside a complex But here's the point. I am allowed to cheat in the simulation I can't stop halfway, but if the end state I get to is physically well defined I am allowed to calculate it that way I Will repeat that in detail after the break because this is an important concept. It's 1033 So let's meet here at just after 11. I'll get started again and Have a second shot at these free energy cycles I think the easiest way might be to take you through an example. It's a bit artificial because I'm going to need to invent numbers But I'll just remind you that there were two reasons we did this One of them was that in general if you have very large molecules Pulling them apart will be computationally difficult. It's going to be expensive Because you're also doing a large change We eventually at some point we want to see the difference between the changes, right? And this is not specific to simulation But any time you end up taking a difference between two large numbers that have roughly the same magnitude It's going to be impossible because that's Well, the day even the difference might be a factor hundred smaller than the numbers But the standard deviation doesn't shrink so you're going to end up with the standard deviation that's roughly the same magnitude as your difference in that case and Just before the break I argue that this might be easier to understand if you had these two large molecules But it's actually going to work equally well if you have one molecule here the receptor E And then you have a small ligand say your drug and this is one type of drug. I prime you can think of if this is I Let's say that I is a benzene molecule and I prime is a Let's say toluene or something So you have one small group added to the molecule So of course I and I prime are different molecules, right? That's there is no experiment that would turn I into I prime directly But in a computer I could take this molecule and gradually disappear this one Gradually get that an atom to disappear and replace it with a hydrogen instead So if you think in the first time in the first case, it's a carbon with three oxygens That's our three hydrogens, and if I gradually turn the charge of those Hydrogens to zero and the charge of the hydrogen I gradually changed that to the charge of a hydrogen and I gradually changed their masses I'm not seeing exactly how we're going to do it, but it's possible to do in a simulation and The end state the starting state is well-defined and the end state is well-defined And what I'm really interested in is that how much does it cost to bind benzene versus? How much does it cost to bind toluene? So let's try to make that difference and the one thing that's important here is to keep track of the signs of the arrows so let's we can use the right example just for fun and Let's assume that the Delta G's here correspond to the arrow that it's closest to if this Delta G is five What is the Delta G gonna be if you follow the arrow in the other direction? Minus five So that it's it's important It's not that Delta G one plus Delta G two plus Delta G three plus Delta G four necessarily sums to zero That depends on the direction of the arrows So but let's make an example here if you first take toluene, sorry Benzene here and then when it's bound in the receptor and turn it to toluene That's Delta G three and I will just for argument's sake say that that is three plus minus 0.1 kilo joule from all So that would be Delta G three and Again the sign matters here. So it's in that direction And then I do the same thing for the toluene here Sorry benzene turning it to toluene, but without when it's not bound. So that would be Delta G four, right? It's exactly the same simulation. I'm doing but now I'm doing it when it's not bound to the protein So then we said Delta G four and let's say that that is three point five plus minus Zero point one kilo joule per mole So these are the raw results I get from the computers But this is not the difference in binding energy These are completely artificial numbers when I do this alchemical operation of turning one molecule to another molecule To actually calculate what the binding energy is we're gonna need to sum up the entire free energy cycle And I would always recommend you when you do this do the sum don't assume that the arrows are placed the same way They were last time So in this case if we start here and we go down Well, then I'm following the arrow in the Delta G one sign So then I'm saying at Delta G one and Then I am in this state and then I continue in that direction That is Delta G four and it's in the direction that I'm following the arrow So it's plus Delta G four And now I am in I prime and then I go up there, but then I'm going against this arrow, right? So it's minus Delta G two minus Delta G two and Then I'm on the top right corner going to the top left corner and then it's against that arrow against it's also minus Delta G three That has to be zero because now I'm back when I started and then I Then I get the equation where it says at the bottom here So if I move over the two negative Let's see me Delta G one and I keep Delta G two on the left hand side and Then I move Delta G three. That's going to change side sign Delta G three and Delta G four that I changed the sign there, too But the point of Delta G three and Delta G four is what I had up here, right? the difference between those two is Delta G three minus Delta G four so that would be Equals to minus zero point five plus minus roughly it's going to be point fifteen kilojoules per mole So the difference Between Delta G one which was In this case, it's actually unbinding Benzine versus unbinding toluene That would be that difference And I never ever did in a simulation when I removed anything. I only made the small changes to any one molecule to the other And this will work equally well if you have two gigantic receptors So this is a trick, but I'm allowed to cheat Because I'm doing things that are easy to do in the computer They would be impossible to do in nature, but at the end I end up with the difference between the two binding things That would be what I could measure in the lab It's not impossible to measure Delta G one and Delta G two But these would likely correspond to taking a taking a simulation and pulling the entire molecule out And then I might as that get fifteen hundred and fourteen ninety nine and The difference between fifteen hundred and forty nine forty ninety nine is plus one But then also because the numbers are a factor hundred larger the errors will also be a factor hundred larger So then the remaining error would be much larger than my difference That is a general result in science never ever You must avoid any experiment where you end up taking differences between two large numbers because you're going to ruin your accuracy Yes Sampling basically because you can't send if you could sample the phase space the partition function completely, right? Then it would be perfect, but any finite simulation. You're not going to sample phase space completely Then there are potential other errors there could be errors in your parameters We will come back to that yesterday. Oh, sorry tomorrow Me I'll come back to that tomorrow when I and I I know that from it was actually last year right where you don't have that much Statistics training so tomorrow I'm gonna have like 10 slides when we talk about statistics in general for life science on a very basic level You know, I'll skip this talking about distributions The point is that if you're calculating a free and it's between two states that are very similar It's a molecule with or without a proton that's not going to change anything else in the system And then you could almost they can If the if the molecules are exactly identical, they're going to behave in the same way Their freedom is the same and then the entropy is the same if the entropy is the same It's really only the enthalpy or the potential difference that matters and then you can calculate that right away But it's not going to work that way in practice for any realistic molecule They're going to bind in different ways and then we'll also change the entropy You will see a little bit about distance on levels for now and you can forget about the things that says grow max here I stole this slides from another presentation. I've been giving In general, we're going to be calculating free engines between different states and Then well one way to do it could be to simulate the entire system and just calculate the populations or the probabilities Because if you know P1 and P2 right then you can deduce what the delta G between them has to be The only problem here is that in many cases, you now have a very large barrier or something So this is going to be exceptionally inefficient The other thing is could be that they might be the free and the difference might be so big that you're going to have 9999 molecules out of 10,000 in the bound state and nothing in the bound bound state and You'd need statistics in both states to be able to calculate relative probabilities, so that doesn't work well either But what we can do is that in physics remember that I said if the left if the starting state is well defined and the end state Is well defined all I need to do is if I can gradually slowly move from one state to another If I can calculate how much energy I'm adding or removing when I do that Then I would know what the free energy between them is and This concept Hamiltonian in physics is really the the description of all the interactions in the system and Normally, I would say so what are all the interaction systems in some sort of state a say when I have when this is a benzene and Then I can also define so what are all the interactions in the system whether since the state B save when it's a toluene and I'm just gradually move between them and one easy way is that all the atoms that don't exist in one state Let's say let's put the charge in interactions and everything to zero and then I'm gradually growing or removing the atoms Completely unrealistic, but it's something I it's well described the equations are well described in the computer and Then I can actually calculate the difference in pre-energy by summing up all the small differences in the energy here while I'm doing it And this is something you're gonna need to believe me. I don't have time to prove this no, okay that Remember that I said two slides ago that if two states are very similar That's so that the entropy is there's that they the states they occupy are gonna be the same in that case you don't have to care about the entropy and What this works on you see the integral there is that at any small point in time We can calculate right here. What is the difference if I move a tiny amount to the left and right and What you can then actually prove is that if you simulate this and calculate this over all possible states in the system So that the average of the change in enthalpy averaged over all possible confirmations Will corresponds to the average of free energy And it's by averaging in all states. That's how you get the entropy part Sorry, that was the part that I tried it's a fairly deep mathematical result for now Don't worry about it. It's something that the program can tell you but we are gradually we're getting the If you have a protein and I it's right now It's 50 percent toluene and 50 percent benzene. I can calculate exactly at this point. What is the difference in free energy? So the way we do this and again, this is not a course of statistical physics I don't expect you to know all the details, but We find some way to describe that Let's introduce a parameter say lambda. So when lambda is zero here Well, the second term is zero so that it's appears and the system would be entirely in the first states So this is where all the interactions correspond to benzene And then when lambda is one then this term would disappear and all the interactions would correspond to toluene And if I'm gradually changing lambda between zero and one, I'm slowly moving along this arrow And no matter if lambda is zero point forty seven All the interactions here are well defined all the interactions here are well defined So I the computer can calculate this it's a lot of equations and you need to evaluate the entire force field But the computer can do it. It's not it's not difficult. It's just that it's a lot of bookkeeping well, there are some problems though that if my atom starts to have a Zero linear jose parameter, but it has plus charts. There are cases where atoms can overlap and everything There are some tricks to get around that that I don't think I'll have time to cover in detail And what you do in practice is that if you have something like this solvent You have this parameter so that the parameter can be between zero and one and typically you tend to pick say ten values Zero is a ten percent twenty percent all the way up to one hundred percent and then you run ten small simulations and Eat simulation only Sample what if I had a little bit more of this molecule or a little bit less of this molecule and Then when I sum up all the changes of all of these I would actually get what it would be the total effect of completely growing my molecule in the water. Yes Okay, good There are some tricks that you might have to do that You don't want these horrible peaks where if two atoms overlap and if I've turned an atom into a ghost They couldn't theory overlap and then I would have an infinite energy But we can see it there, too So while I'm in the middle of this process I say that oh I allowed the atoms to overlap the energy is not infinite even if you overlap as Long as in the starting and end states I move back to something. That's physical and realistic If you didn't follow that don't worry for now But my point is this that this is what you get to so if this is This is an alcohol that I'm salivating in water And then the lambda parameter here goes from zero to one and in this case is more is probably 20 simulations And in each of these points I then calculate roughly how much is the free energy changing here as a function of lambda So if I had in this particular case if I had a little bit more of the molecule It would not like it because the change here is positive while here. It's negative So you end up with a bit of variation and the variation here comes from as I'm turning charges on or off Or if I'm turning the Lena-Jones parameters on or off or if I'm turning the bonds on or off So it's not you can't really understand exactly what happens in these curves But the point is that then I just take the integral under these curves and the integral under this curve will correspond to the total difference in free energy and As noisy as these curves look the integration operation is averaging So you actually end up with fairly good values of Delta G Computers today for small molecules computers are at least as good as experiments at calculating the salvation free energy as small molecules And this is something that is used on massive scale in the pharmaceutical industry You get salvation of a new molecule in 10 minutes rather than It's not so much that it's difficult to measure in the lab But the part that's expensive is of course you need to produce the molecule to measure it in the lab. Yes So this had to do there are slightly different ways this has to do with trying to avoid those peaks The red curve has to do it if we do it in vacuum and the blue Her blue and black has to do whether we do it in water So these are actually just slightly different ways of calculating it and they might look like they're completely different Right to the point is that the integral of them is going to be roughly the same So they take slightly different paths in this non-physical space between the two states But the final value when you integrate them is going to be roughly the same So here the blue curve is lower, but there it's higher So the total area under them is going to be roughly the same if you thought that was difficult There are easier ways to think about free energy Potentials of mean force is something that we hear quite a lot And I already showed you one of those and the potential of mean force is exactly but it In general the force is the derivative of the potential right So if you think about potential energy or something the difference in energy between two states The derivative of that is I'm changing it is the force I need to apply to it But if when I took that molecule and pulled them if I measure the force and if I slowly measure the force if I'm pulling something If I'm measuring the force I Can effectively get that potential back And what I said before the break is that yes if I measure the potential all the way when I'm pulling something out I can calculate how much it costs to pull it out But it's not just a matter of pulling it out I actually get this as a function of say the position if I take a small molecule in water And then I gradually pull it into a membrane I will actually get the entire shape of the molecule that how expensive it is how it water How good is it to have it in the head groups? How bad is it going to be to have it in the center of the membrane so I can get this as a function of the position? Or if I have a large protein the iron channel I showed you as the iron channel is opening or closing right I can actually trace the potential free energy Not just in the opening closed states, but I will see what is the transition state How much does it cost if I'm as I'm forcing the molecule or something to follow a particular path? What is the energy at each specific point? So that the force is incredibly noisy But the average of the force in a simulation if I'm under equilibrium if I take the integral of that That's going to be the free energy with respect to the coordinates that I'm changing Let's show you an example of this instead If I take two molecules one molecule here and one molecule there if I Define some distance between them I can say what is the energy as a function of this distance and then I run this in a simulation And then I change this distance. I force and tie to be further apart or closer together And the way I do that is literally by attaching a small spring with them And I can get the spring tie to pull them together push them apart Then I will see what is the energy as a function of distance between these two molecules So at each distance, they're going to be ton of different relative orientations and everything so that I will need to sample I can't just calculate it once and The really important part here has to do with a membrane I can take a molecule and gradually pull this into a membrane now if I just calculate it for one frame here one snapshot They're going to be noise the atoms here with us at each point here I need to simulate a few tenth those tens of thousands of steps or something so that I average all the properties of the Molecules and effectively get the free energy instead of just the energy But if I do that I will literally get a curve that describes how much any do I gain or lose at each z-coordinate here through the membrane Yes Do you mean lambda or yeah? Yeah, lambda lambda is just a what you call a parameter, right? So that we we use this parameter to describe as I'm going from one state not so here in this case We typically don't use lambda, but you can say that lambda is zero here and one here There's just a way to describe a relative change. It's a dimension less parameter So just so that we can translate this to the computer But in this particular case, it's just a matter of measuring how much force do I need to apply to this atom to keep it at this position and Here if this is water-soluble the force will likely be roughly zero, right? The molecule will likely move up and down a bit so that it's going to fluctuate a bit But in general will be zero But let's say that this was a hydrophilic molecule and if I now move it into the membrane and I put it here There's going to be a gigantic force that the molecule would like to move out of the membrane So I have to apply a lot of force inwards to keep it in the membrane But this too is going to be noisy. It's going to fluctuate up or down But in general I'm going to need to apply a force downwards to keep it here and if you integrate this force You're going to get a free energy. Let's see here. This is what I said hydrophilic, right? So it's going to be like to be out in water and then it's going to hate to be in the membrane and then it's going to like to be out in water probably and Can you imagine doing something like that in the lab calculating how expensive it is to keep a specific helix say in the head group region? I Can't even imagine an experiment. They can do that so that the neat thing here Is that we can use computer simulations to understand processes even understanding transitional states? What if this is a horribly if this is a very hydrophilic molecule Somewhere in here at the very peak of the membrane, right? It might even look something like that That is by far the worst possible case. You can have this water-loving molecule But that's kind of like a transitional state. Could you imagine that that's important? well Understanding whether arginine would like to be in the middle of a membrane We don't need a computer simulation to tell us that's Not very likely, but we have transport through membranes So how frequently is how easy it is for a water molecule to go through a membrane? If you think a lipid how frequent is for a lipid to flip from one side of the membrane to another side of the membrane? We know that it happens in nature, but it's a very slow process The speed by which this happens is going to depend on the transition states, right? And what had we just said about transition states the last few days we can't observe them The computer can observe So in the computer I can force the molecule to be at the transition state and Measure how expensive is it if I force you to be at the transition state? I know you don't want to be there But if I force you to be there anyway, I can measure the barrier of the transition state in a computer Which is completely impossible in the lab So sometimes there are there are two reasons free engines in some cases We it's some cases that we want to do it on massive scale, and then it should be cheap and fast Hopefully accurate too, but primarily cheap and fast because it's so expensive to produce the molecules While in other cases is more that you would like to understand a process that it's completely impossible to study in the lab And in this particular case, we might just be interested in one curve I don't really care if this takes one month to calculate Because I can't do it in the lab anyway The other thing that's important with free energy is that they are strictly reversible If I take a molecule and disappear it Turn it into a ghost or if I'm starting from a ghost and gradually growing the molecule in that place They have to be the opposite sign If they were not the opposite sign, I would violate energy conservation and everything The same process in different directions only the sign can differ So anytime you calculate free engines, whether it's in a computer simulation or in an experiment Check there is this famous quote in politics right trust but verify Do your positive and negative controls you know that it should be The opposite so let's say that I calculate the free energy of binding and I get two results First I calculate the free energy of binding and that is minus three plus minus one kilo joule per mole And then I calculate the energy of unbinding and That is plus nine plus minus two kilo joules per mole What happened? We don't know right, but there is something wrong here because that's not possible Maybe we made an error in the calculations It's sadly, it's maybe we completely underestimated these errors for whatever reason, but there is something here. That's wrong They should be the opposite sign if they are on the other end if you see that that's plus three plus minus two then it's fine And again, don't this is not specific to simulations if you can measure the same process two ways do it because that's your safety check that you didn't make a mistake the first time you did and that Leads to the other things that we're going to talk more about tomorrow Estimating errors you freak. Do you think it's important to be accurate in science? No, then we wouldn't use bioinformatics As it's not be dissing bioinformatics You think about bioinformatics bioinformatics is full of errors, right? Every single sequence alignment you've seen isn't correct So why does bioinformatics work? Well, that depends It's dangerous because better than nothing sounds good So would you be willing to take a drug that was developed with my better than nothing technique? So you have high the I have this new drug and let's say it's against headaches And it's better than nothing only one out of four patients die Would you like to take it? Okay One in four billion patients die So the point is It's not that you need to be accurate, but we need to know how good or bad we are So you can make an informed decision and In some cases that maybe one in ten is good enough If it's just a side effect that you get a bit tired from the drug dying I would probably prefer one in four billion So that you need to know how good or bad you are the point is yes, of course all other things equal It's always better to be more accurate and precise than not being it, but you don't have that choice So the point with bifurma example you have the p-values. We know what is the likelihood of this model being wrong? And then you can decide if it's 50 50 you likely shouldn't submit it to science But we're not saying that the model can't be wrong. It's just saying it's in this particular case It's not very likely to be wrong and Then what what error you accept that will depend on the scenario So I'm going to talk about that tomorrow and what errors are in particular Estimating errors is super difficult and there are a bunch of very advanced techniques to do this So if you have an entire series of values The obvious one is to just calculate the standard deviations has turned that into a standard error And if you didn't follow a word of that I will cover that in detail But there are ways so based on the fluctuations and the data I can estimate Not just the standard deviation because the standard deviation just measures how much my data is fluctuating There are ways from this to derive roughly how good is my prediction and I will cover that in detail tomorrow But there is a problem here Let's assume that I'm drawing a curve for you. You've all heard of standard deviations, right? So let's say I'm plotting a function here. What is this? If I now give you an estimate of the Average is this Sorry, this is some sort of value. Why here? Do you think this is a reasonable estimate of the Average somewhere there, right? Unfortunately, this was the temperature in Sweden and I only measure it from January 5th to January 10 So we just predicted that the average temperature in Sweden is minus 10 degrees So the problem is that the temperature January 6th is somewhat correlated with the temperature January 5th, right? and As you simulate this over an entire year or something eventually the temperature will start to pick up The point is that we don't have 500 independent samples. Each sample is highly correlated with the previous sample And this is something that you you you we usually sweep that under the rug in the beginning courses in statistics There are ways to treat this too There are a ton of advanced statistical methods that you can use and this of course you have to know for the exam The pointed A there are programs to do this for you. Gromax our program is one of them But in general what do you do when you have a problem like this? First you start to look it up properly Don't assume that you know statistics Ask a statistician if it's very important. Remember those things that I told you earlier on. No We're gonna have that tomorrow. Ah p-value hacking. There are some really fun There are a number of very embarrassing papers including from groups from Stockholm that they have published amazing results And it's just because they didn't understand proper undergraduate statistics But undergraduate statistics is easy if your undergraduate was in statistics So how many of you have an undergraduate in mathematical statistics? So this is you need to know that you don't understand this Even I don't understand this properly I have a bit more experience than you but there is a point here where you need to ask a statistician statistics is Dangerous and difficult it's difficult But it's also the big danger here is that you think your results are better than they really are and it usually has to do With these correlations that I showed you don't assume that you can just apply the standard deviation or anything you learn in undergraduate statistics, but Quite a few programs that you use either in the labs or later on They will have a way to if this is my data they will estimate roughly how accurate your data is trust those programs and Anytime you get a number from a colleague or something unless they have Here's my result a Result that looks like that with one bar It's useless completely useless Because there are two alternatives here right one of them is that the standard error is like that The other one is that the standard error is like that and in the first case It's a very important result in the second case. It doesn't tell you anything. You don't even know what the sign is Unless there are standard unless there are error bars on numbers. They're worthless completely And the whole point yes, so this base when I show you that this is an example of the energy in a water simulation and the point is that Saying that Delta G is minus 15 doesn't tell us anything Minus 15 plus minus 1 tells us something we know that it's negative and we know roughly what the error is and The error the best you can hope for either in an experiment in a simulation is likely going to be a few kilojoules per More it's very hard to get free and is more accurate than that Having said that we have calculated this a ton of times both in simulations and experiments and everything These are Anna Anna Johansson or it is a decade ago time flies So all those Biological hydrofibicity scales we've been able to understand them and confirm them just by putting not even entire helices But just the amino acid side chains in different parts of the membrane and measure how expensive it is And they can say for instance arginine. We see that it's good to have it in the water It's actually good to have it in the head groups too, but it's bad to have arginine in the center of the membrane and All these curves we can show that they correlate well with bioinformatics and everything So it's surprising the simple simulations. We can understand fairly complicated biology Do you see alanine on the other hand alanine is happy everywhere? Both in the center of the membrane and in the head groups and you can do this as Complicated as you want you can literally take an entire helix and pull it in through a translocom and measure the force as You're pushing the helix out in the membrane people have done this since Basically your your imagination is the limit here Can you try to imagine an experiment that would measure the transition state as a helix is passing out of a translocom? It's completely impossible and simulations can do it when I was your age We couldn't even dream of it today. It might take you a month Which is certainly you can't do it for a thousand sequences, but you can do it for a few to help us understand So a lot of the things that I've hand-waved about then everything are actually based on that type of calculations It might sound strange to have the study questions already here I'm not gonna let you go quite yet We will talk about these tomorrow But to prepare a little bit for this afternoon and not to leave all this strange theoretical free and just stuff hanging I'm gonna show you a little bit. How about we use it in the lab? So I might not cover all these slides in detail But this is a research team that has been going on in my team for close to decay now So I told you about various types of iron channels One of the types of iron channels that we are most interested in are the ligand gated ones So these are the ones that sit between nerve cells and when you get to the end of one nerve cell you release on neurotransmitters These neurotransmitters diffuse over one millimeter or so and Then depending on what neurotransmitter you have they will bind to the next nerve cell and then create a new nerve impulse I even think I have a Now we're gonna and what I love here is so this is kind of halfway between physics kept this You probably you're too young. This is Sydney Harris for a long time He made this beautiful cartoons and scientific America's jokes about science And I think this had he's retired since over a decade So where all this started is actually long before we had iron channels There are a bunch of concepts that this is a fresco from Babylonia 4,000 years old Where they're showing people smoking or drinking alcohol And alcohol is fascinating because it's it's a drug that it's probably the by far the oldest drug We've had so we've used it and abused it for for millennia Nicotine is another example. Benzedite subpoenas Rock stars what all these drugs have in common is that they actually hit the nervous system And what I show it here at the very end here. This is pop of all which is an anesthetic Which he used to sedate people in surgery So that's quite serious business. I think I might have showed you the slides here, right? That the first charges you could make this would not be possible without the drugs at the time We had no idea what they were actually doing But that we do know today So that what we are interested in specifically the blue channels that we had on here What happens when something binds here and what is that controls this to some sort of allosteric modulation? And I think I will show you this slide already, right? that We know that there is something happening here that we're binding or changing this channel Depending what I'm binding here and depending what channel I'm binding it to they're either going to open or close So it's exactly the type of process. We're ideally we would like to design drugs I'm not interested in running a pharmaceutical company, but we would like to understand what happens and when we control it What happens on Friday if you have a glass of alcohol? Why do you get this warm fuzzy feeling and it's These are entirely allosteric modulators. They work just like transistors a small change in the binding here will some are create a large change in your nerve impulse and For decades here the entire field was fumbling in the dark because this was about as good a structure We had of them and then the last 10 years things have exploded We have a number of channels both from bacteria worms Then increasing those from humans and mice so suddenly Can you imagine the difference between the previous slide and suddenly we know every single binding site in these channels We have a ton of information. So now we with these structures. You can start designing drugs So already when we got the first bacterial channels, I had a very skilled French postdoc And he could even run simulations and show specifically where in these channels between these healers is that various anesthetics and alcohol drugs bound and they could also show how long they take he Could show the specific hydrogen bonding patterns And at this point we were super excited because this you probably can't see it But this is s267 there It's a serine and this serine residues has been shown to be super important if we remove that we completely changed the sensitivity to anesthetics and alcohol So the fact that in a simulation we spontaneously see the molecules go there and bind if we just add ethanol out in the water So this is from our point of view pretty sexy. We felt that we had understood that the binding sites was we can It also made a whole lot of sense It's kind of like pushing your foot through the door that we're pushing the subunits of this channel apart And when we push them apart that would somehow explain why the channel would open more And we even published this We were pretty happy and then we went down to Paris and I talked to some colleagues at the conference And they just pushed out a new structure and this is an x-ray structure and it's a type of x-ray structure Yes, you call a co-crystal which is That's you have the extra structure, but you actually have a small drug bound and I think it's desk chlorine in this case Or no, it's pop of all The mesh you see here is the electron density. You don't see the atoms But again, this is a very good electron density. So it's obvious that there is some drug bond there There was only one small problem here and it's that I had bet the bank All our simulations said that the drug would bind there So this was me before the talk and that's after now The reason why I'm showing you this that we made a bunch of mistakes here And this is a bit of memento more that all the simulations here were done in the bacterial channel There were homology models. We didn't have much choice That's right that the simulations were done on prokaryotes The experiments was done on a different type of channel and everything so that it's we kind of tried to cover too much So I used the bacterial channel to try to make simulations predictions about a human channel the glycine receptor Which is important for alcohol binding while all the structure was for the bacterial one So at that point we actually took a step back and we started doing quite a lot of experiments You will see Stephanie actually and maybe harsha Talk a little bit about their work and what we then do is so we take these small frog eggs Where we express the channels and they will take you through it this afternoon and that way we can actually Specifically measure in the lab. What happens if I add a bit of alcohol to this channel? Does it open more or less? What if I make a mutation in the position 267 do I get more or less current and The way research happens right is that it's not by one study, but you need to make dozens or even a hundred studies So you had lots of statistics and see How do things change depending on the mutations and to make a long story short here? What we actually discovered is that there are some pretty important differences between the prokaryotic the bacterial channels and the human ones And in particular some of these binding sites that we had more or less by coincidence found They're super sensitive to a mutation and this is a mutation that in the bacterium You would not have this binding site been in the human channel. We would likely have it So it's at this point. It was somewhat involuntary that we had likely found a binding site That is present in the human channels, but it's not present in the bacterial ones And then we took those lab results and went back into the computers And this is when Samuel started to do a lot of mutations and everything And we actually turned out that then the simulations to if we compare the bacterial channels to the human channels There are two binding sites. There's an orange binding site and there's a purple binding site And in the bacterial channel by default you would only have the red binding site But when I make the bacterial channel look like a human channel, do you see what happens? I open up a new binding site So then we would have two places where the iron can bind and you can actually test that in the lab too This is all the results are very bad statistics in the wild type. You had binding in a few places In the human one in the bacterial one we see very good binding you might not see it there But they're binding in the opposite places actually and if I take the bacterial channel and make it look like it's human Then I actually recover the binding that I had in these human channels And so we're here We started to be a bit more ecstatic because it actually if you look at the biology of these channels Which in hindsight is a bit embarrassing. We should have done that were much earlier It turns out that the bacterial channels are mostly shut off by at least longer chain anesthetics By the human channels they're rather turned on So they have this roughly the same channel. They are homologous and If you have a very short chain alcohol, let's see if I have that in the previous slide Sorry Don't take it literally Small alcohols methanol and ethanol they are Even the bacterial channels are opened by them and what happens with the longer chains alcohol some you get the opposite effect So if you what is the difference between ethanol and proper for a propanol is just one ethyl group CH2 And instead of opening the channel I close the channel instead and The human channels on the other hand everything up to hexanol will open the channel, but the longer ones will close it So that the hypothesis that we then developed is that it's likely based on the size of these cavities So that one of these cavities the red one should then be closing the channel Well, if you bind in the purple one you should open the channel And we managed to publish this and it actually it wasn't strictly true But it's largely correct. There are definitely separate binding sites and depending on the specific channel We have we can open one or the other The memento Mori here, too, is that just this entire department we love to work with prokaryotes prokaryotes outstanding They're great model systems. They frequently have roughly the same properties I can overexpress any prokaryotic Protein in a matter of the final days but weeks at least it's easy to express prokaryotic channels You carry odds everything is in order of magnitude more complicated So it's obvious to make the assumptions that use the your carats as model systems But there is one assumption there start by checking that they actually behave the right in the same way Because this is not limited to these and there are tons of channels that don't behave the same way in prokaryotes and you care But at this point we still had this nagging feeling that this was a bit of hand-waving and everything What made us super happy though was after we had published these studies About a year later There was another channel. It's a much higher organism. It's essentially a human. Okay, it's a worm But it compared to bacterium is essentially a human to first order approximation and this It had a molecule bound. So this is ivermectin which in the antiparasitic agents and This molecule is bound in exactly the same place where we had predicted that Eukaryotic channels could bind things and the point is that we had made these predictions before they published the structures So I'm not saying that the experience simulations are always right But there are increasingly we are increasingly doing our research in the computers and then confirming them in the lab And at this point we wanted to show that we actually do proper prediction So what we would like to do? Can we take one of these channels and can we take a channel that is normally say turned off by a molecule and Change the channel itself so that the same channel is instead enabled by the anesthetic so what This is a to be this drawing if I recall correctly small anesthetic volatile and If we take a wild type of the bacterial channels, we know that these channels would be turned off by the anesthetic And we had a rough idea where some of them bind because there are crystal structures of this flooring But let's not assume that too much Remember what I said so why on earth would I this is just a wild type the crystal structure? I know that things bind there, but I wanted to test these two different binding sites There was one binding site inside each subunit and one binding site between the subunits so the one inside and we called intra subunit and the one between is inter subunit and The one inside the subunit. This is where I have things bound in the crystal structure even so why on earth would I dock? Just something where I know the result So here we literally did high throughput screening, but just one molecule. I was where does destroyer in bind So why would I even ask where destroyer in binds if I know where it binds? So this is my positive test right if I had predicted in this case that this roaring binds here There is something wrong because I know that it should bind there And I get it roughly right I think there was one of this group that's flooring was the turned maybe a quarter of a turner something But for all intents and purposes, it's awesome I do predict that it will bind where I know that it will bind which gives me some Reasonable confidence that the docking should be able to capture the properties both of my entire protein in a membrane It's a fairly complicated setup docking in membrane proteins is far more complicated than docking a water-soluble protein Which is why most people haven't done it and Then I also take the same molecule But force to dock it between these subunits and you can almost see that here that There's a large phenylalanine ring in this case and they're going to overlap That's going to be a very bad any they will hate to be there And then I take exactly the same molecule but mutate that phenylalanine ring to something small like an alanine I literally remove the entire ring and Then I do the same thing here. I dock it and say that What are the positions you can bind in inside each subunit and what are the positions you can bind in between the subunits? That doesn't tell me anything Because here is the problem with docking What docking is going to tell you is that yes, you can bind there if you're going to bind here You're going to bind in that post if you're going to bind there you can bind in that post But that appears to be pretty bad Here you can bind either there or there So docking basically just says yes or no and we said yes. No. Yes. Yes. It's yes better than yes Or is the third yes better than the second yes Yes is better than no right, but it's only it's unlikely to be in the no state How much better is yes than no and this is where docking fails docking is not going to get to a Delta G So it's I'm not saying that no does not mean that it will never happen I will eat my left shoe just no just means that it's not quite as good as yes and Quite as good as that does that mean one in ten one in a hundred or one in a billion and To answer that we need to calculate the free energy But the the reason why you need the docking is that now we have a post we have four poses We have the wild type Inside and between subunits and we have the mutant inside and between subunits. So there are four things I want to test Oops, sorry And I'm not going to go through the details exactly how you do that in a simulation There is lots of data on the spot, but it's fairly let's We use exactly the thing I told you about before the break We calculate what is the free energy of binding desk floor rain in a specific binding pose between the subunits in the wild type and Then I do this I Said between our so the other one would be inside So there are four different things I wanted to practice is much more than four But I'm I will ignore that for now look at the let's start by looking at the black bars there ignore pretend You don't see the orange ones So this is a molecule called this floor rain and we're only going to look at those two black bars and Negative is good binding so that this roughly minus 22 kilojoules per mole. Do you see the arrow bars? Do you trust the results? So the arrow bars are reasonably smaller, right? Now there are some caveats here those error bars only describe The sampling when I calculate my free energy. They don't describe the quality of the homology model So there are you should always be aware that any error bar almost any error bars from any calculations They are lower bounds. They can always they can only be larger in practice because there there are always errors We don't account for or the quality of the parameters or so So which place is better to bind in inside the subunit or between the subunits? So it's gonna be better, but what's the difference of minus 22 kilojoules per mole and say minus 14 kilojoules per mole The point is that both of them are good, right? So you're gonna see binding in both places So if you start by adding one molecule, you're first going to bind them There are five sub there are five subunits so that you will first add five molecules here And then eventually when you've saturated this binding site you might start to bind there, too So both of them are good, but this one is better The relative occupancy of these two states you can calculate from the Boltzmann distribution It's that simple But you will definitely I wouldn't do that in this case, but so we definitely prefer to bind Inside each subunit that made sense Because that's what we saw in the wild type that if you made the co-crystal We found the anesthetic bound inside each subunit So it and again check check check Trust but verify so we verified that we predicted the place where we predicted it would be best It's also the place where we saw it when colleagues of us determined a structure and that also agreed with our hypothesis That this binding site inside the subunit should be the binding site that shut off shut the channel off That we would inhibit the channel from binding them Everything makes sense But then we look at the orange bars so the orange bars is exactly the same binding in the wild type Do you see what has happened? I don't change that binding that much, right? It's a little bit worse But this binding is becoming way better and Now the problem is that the difference between the two error orange bars here are smaller But it's certainly outside of the standard error estimates there So what do we predict would happen in the wild type? As we're not wild type a mutant my bad So what we're saying in this f238 a mutant We would predict that they would start by binding between the subunits instead of inside each subunit and Our hypothesis is that that that binding site would have the Opposite property right that it would open the channel instead and then we the problem with doing this for one molecule It's noisy so we actually tried another molecule chloroform here, too Which is another small anesthetic and qualitative actually the results are even stronger for chloroform But you see the qualitatively the same effect That's also a very common point in science that don't just use one molecule do it for five or ten molecules and look at the patterns So based on these results, this was entirely computational. What do you predict would happen in the lab? So I said in the wild type if I added this anesthetic I shut off the channel So what would happen in the mutant now if we add anesthetic to the mutant channel it would open it and it does So that all these different and here's the point you see there are quite a lot of fluctuations for different types of anesthetics here But for all of them in the mutant they turn on the channel instead So you reverse the problem It's even a super strong effect so that it's when you go for lower concentration You get a super strong opening instead of inhibition Real life is more complicated than that because if you thought it was complicated with two binding sites We actually there is a third binding site You can also bind in the pore and part of the inhibition actually happens in the pore So and to make it more complicated than this all these channels. They're also they have bimodal effects meaning that Normally if you add a little bit of something and you get an effect if I had twice as much chemical You would expect the effect to be twice as strong, right? Or you would affect to be stronger at least what frequently happens with these channels is that The effect the sign of the effect is concentration dependent So if you give it a little low concentration or something you turn it on and then as you increase the concentration Suddenly the effect starts to go down and if you increase the concentration even more you shut off the channel So it's not just the slope of the curve that varies, but even the sign will vary dependent on the concentration Can you guess why that happens? Sorry this can you guess why that happens if you look at these curves. Yes So what might happen here, right? Did you start to bind in this black site and that is definitely the best site And that's the site that would turn the channel actually the yellow this one might be better. So you would start by binding Between the subunits and that would open the channel But you only have five such binding sites and at some point in time all those five sites are occupied And as you add even more anesthetic The anesthetic will also start to bind inside each subunit But the effect here will be the opposite So now as you increase in the concentration even more you start to take the effect down and at some point This might even be a stronger Most of you have probably tried that at some point. So what happens where you take one glass of wine? Or two you feel a bit aroused, right? If you start drinking ten glasses of wine You become tired So that these effects are concentration dependence and alcohol is included here It's slightly more complicated for anesthetics and everything but that this is one of the reasons why neuropharmacology is so cool It's all these the normal things about binding and everything don't apply it multiple binding sites. It's concentration dependent It's super complicated. So there is like there's room for 50 more carriers. I have The other things you can do with these potential of mean force You can actually measure how expensive it is to pull an ion through the science channel And the reason for that is that then we measured the transition state here, right? Then we can measure What how quickly will ions be able to move through the channel? And then you can do this once if the channel does not have anything bound under when we expect it to be closed And then we have a fairly large barrier and then we add The glutamate the neurotransmitter here And then we would expect the channel to be open and you can actually show that Once you have an open channel with this bound there is still a barrier, but it's much lower So then the channel is actually open So we use computers quite actually actually to predict what happens will the channel open and close and suddenly it will conduct ions If you thought that was complicated, we just worked on one specific channel here That was also a homo pentamer. So it's five identical subunits. I might have mentioned this to you But I didn't show you the slide One of the most famous channels in the nerve system is the GABA channel And we can even there have been a few structures of this and then it's been a homo pentamer The important GABA channel has both alpha and beta subunits and again I think I mentioned this to you that the interface from beta to alpha This is where the anesthetics will bind to sedate you If you change this binding site, you can create mice that you can't sedate with anesthetics If you think that was complicated the most common one in your bodies is actually has both alpha beta and the gamma subunits The order by which you assemble this is also going to matter If you think that was complicated, there are delta subunits in the channel too, they're not that common, but they definitely occur These are just the major subunits for each of these subunits you have different genotypes So there are like six slightly different genes for the alpha subunits Three different ones for the beta, three different ones for gamma and then delta, epsilon, theta and phi subunits All of these genes are expressed in your brains And exactly why and how we don't know that there are some very weak sequence differences between them There are small things, they're of course important Otherwise you wouldn't have 17 different genes, but exactly what they do your bet is as good as mine Really interesting research here for the future generations both in sequencing actually one thing I'm done, but I'm just gonna say one thing one thing that We think should happen that but that we don't know is that when it comes to addiction disease in general It may somehow make sense that the expression level the body uses should vary here So that what if normally this channel is six normally you don't have that much ethanol in your bodies saying and not a molar or something And then you have a stable normal cellular response Remember that I mentioned when the ethanol is present you tend to stimulate the response so we get a stronger response So what if you regularly drink alcohol so that you always have 100 millimolar 100 millimolar you would be dead, but But you always have one millimolar alcohol in your blood Eventually the cells will likely start to downregulate these genes Because you have too much response you don't want that much response and you have the alcohol taking care of it anyway So then your cells will likely downregulate certain genes here and Things are gonna work just fine As long as you continue to drink lots of alcohol And we don't know that I wish that we could prove this but we haven't So there have been a couple of large scale studies where people are trying to find differences in expression patterns and that has to do with transcriptomics And in particular looking at people with addiction disease do they have different expression patterns and I'm it's inconclusive this far But the reason why this is important is that we know today that many of these withdrawal syndromes They're physical we historically we've thought of them as psychic Relatives and but it's not that you can you can actually die from withdrawal syndromes if they are severe enough So what likely happens is that you are in this situation where you've downregulated all these teams and suddenly you stop drinking alcohol Which of course long-term is very good for you But short-term you have a nervous system response that is only one-tenth of what it should be and You usually you're gonna be physically ill from this Not from not thinking alcohol right because what your nerve stones respond the way they should and then of course Eventually if you do the slowly will yourselves will hopefully up regulate this again And what we then think but again We don't know likely the difference in expression between all these sub genes is likely what experience We have different cells in different parts of your brain in your spine and your hypothalamus and everything But we have no idea about the details And this starts going and being related to brain Research but the brain and neuro pharmacology and neurochemistries in my somewhat and biased point of view Some of the coolest research. We're going to do the next ten years Three minutes past noon. Let's finish there and then we'll meet those of you who want at 2 p.m. At sideline flap