 It's nice to see you again. I hope you had a few fun weeks with Burke and Lucy and a bit of time off What's the week ago? What we were traveling in the States? Today, I am going to go to two or even three things The main topic for today is drug discovery and docking which is a concept that's not really covered in the book at all There are lots of things that have happened here I'm gonna start though with doing a bit of recap of the folding because Lucy told me She didn't have time to go through all the details about folding units and Some of the stuff on cold denaturation last week So I will cover that a bit in particular because originally I had a few slides about deep protein coupled receptors here But I know that Lucy already covered that in the membrane protein lecture So I think we should be fine in the balance of time and That also provides a good way to reconnect with last week And then I'm going to talk a little bit about docking and the entire drug discovery pipeline in practice And at the end I think I'll have a chance to share some of the very cool stuff that has happened recently with the coronavirus But I will get started and talking about folding in general So I think Lucy told you a little bit about Christian Ampelsen and his result and I know I've covered this before And it's one and I know that I told you but I need to remind you that it's one of those results That seems so completely obvious in hindsight that proteins just throw it in a glass of water Or a test tube and proteins you can denaturate them with high temperature salt concentration or something Or pH and if you reverse these conditions, then the proteins should spontaneously refold should is the keyboard There are some examples here that is difficult to unboil an egg But small proteins in general will be able to refold This was so not obvious in the 1940s and 50s and that's of course the reason why Christian Ampelsen eventually got the Nobel Prize for this But today we take it for granted and it's one of those things we don't even talk about it in the course literature because it's obvious But The nature of this transition is not obvious And we started by covering this already at the very beginning of the course when we talked about all these transitions and we Made the parallel to phase transitions where there's gradual or all versus non transitions and then we spent several lectures on that I'm not gonna go through all that again But in the case of folding there are two things that are important first the extent to which semi folded states exist and we talked a little bit about folding intermediates early in the class and The other part is that what is it really that folds? Do you start you fold one protein or do you fold part of a protein? Or do you even need several molecules together for them to be able to fold? And we're gonna go through one example today that really helps us understand What is that folds in a way that I find kind of beautiful? There are many ways to measure this and the challenge here that We somehow want to measure how much energy or ideally I would like to see the movie and practice that what is it folding? Apart from computer simulations, which are models We can't really do that though. We're gonna need to access the seed directly And the classical experiment that you would do in a physics lab here is calorimetry and calorimetry is a Really simple technique just measure the amount of heat we're putting into a system The slight complication there is that the amount of heat we need to put into full proteins is very limited So that you need a very good insulation and you need to be able to measure very precisely Exactly what was the change in well temperature specific heat and how much energy we need to use for it And the way you typically determine protein folding or unfolding in this case is that as we are Adding heat here and as the temperature goes up Eventually you see some sort of jump in the heat capacity here And then when we get down back down here the heat capacity has changed So there is some sort of offset So change in heat capacity is usually a very strong indication that there has been a phase transition or something It's also it's not really abrupt in the sense that it occurs over Instantly at say it well compared to exact 100 degrees of boiling But that was related to the things we also brought up at phase transitions, right? That on a microscopic scales when we're looking at individual molecules It's reasonable that this change actually happens over a limited temperature span a few 10 degrees So what this tells is that there is some sort of abrupt change There is nothing that happens down here And then there is some sort of temperature interval Over which something happens and in this case, it's the protein unfolding But the mere fact that this happens is a particular temperature Does it that doesn't really tell us anything about the cooperativity or anything? It doesn't tell us whether it's part of the protein unfolding one folding unit Or if you need 10 proteins together, it just means that it happens fastest at temperature So we're not really that much wiser just from looking at this plot But then people had thought about this and then they thought Lots about it So based on this if you know how much protein there is in this particular test tube and we know how much energy we're putting in Well, then we know what is the folding or melting energy. That's the same thing different signs Per unit of protein per molecule, right because we can count the molecules And then we're going to need to find some sort of way to measure How much is the energy per melting unit or per folding unit? We don't know that yet But assuming that we had such a method then we could start to compare that If the amount of energy Sorry, if the amount of energy per melting unit is the same thing as one whole protein Which we get from the previous plot Then it's an all or none of this says that you need one protein molecule at the time folding If the melting energy or the energy per you folding unit is less than the full protein That means that the folding happens as part of the protein And in theory if the energy per melting or folding unit is larger than the full protein That would need an aggregate of many The problem is that it's not at all trivial to find out what the energy is per melting unit So i'm going to take you through a smart trick in terms of equations that helps us get that And it's also one i'm going to there is no solution for me But to show you this and how this is derived But again, this is not obvious and if I had not thought about this I would not be able to spontaneously come up with it I would rather waste 10 hours of trial and error paper first So We're gonna need to talk think a little bit about energy and entropy both in the native states and the molten states And that goes back to the thing that I tried to hammer in in the early parts of the classes Anytime you get a question about a process or you need to think about a change You need to separate before and after don't start to think that oh the free energy is higher The free energy is slow. It's almost a matter. What is it state a relative to state b? So here we call the native states E and s and the molten state then we just add a prime and that means that we can define the free energy in each of these states And Well, if we simplify this to say that there are only two states. There is a full native states and there's a molten state Then I can write out the partition function because the partition function only consists of two So there's one that is the one without prime and then there's one with prime, right? I like those type of partition functions. It's almost it's just as easy as it was for you in the lab And this actually shows that a simple and horrible as lab one was You don't necessarily need more than two states to start to capture interesting things And then if I want to say, okay, so what is the likelihood the fraction that should be in the molten state? Well, that should be the free energy of the molten state divided by that sum, right? And then I can simplify this a bit by taking e prime minus e And then all these turns out into one divided by one plus And then the change so just dividing everywhere by that expression Do you follow that even if you didn't derive it yourselves? So it's just basically I'm just written out the probability of being in the molten state For a molecule And the key thing here for a molecule Because this should be the melted unit or so Then you'll also see that there is some sort of energy as a function of a there is some sort of regime here Where things happen and that's what we can measure. That's what we can see in this calorimetry There will be some range over energy or sort of a temperature when the melting happens And the exact shape of this curve is super difficult But the cool thing this happens over a very relatively narrow regime So when we're starting here, we have not molten the molecule at all and when we finish the entire molecule is molten So the first approximation that we can say well the entire molecule everything melts So that the fraction in molting goes from zero to one And that's the entire temperature span over which the folding happens So it's just the simplest possible approximation of the derivative here. Everything changes roughly over the temperature range I have And you can argue that it's horrible, but I'm interested in first-order approximations here That's sadly an art that is lost because you have so many computers and calculators and everything that don't start doing things Advanced do it with paper and pen But we also know so this is something that I can calculate from that Graph I had but we also have an expression for P molten on the previous slide, right So what you can do we can calculate this derivative to from that long expression I had so now and that should be the derivative of the long expression And if you know your exponential laws and derivatives there that will fall out to be this expression Either you do it yourself or you just trust me that that's the expression It's just derivative of logarithms with respect to temperature And then this small term falls out due to the chain rules and the derivatives in side the exponent And if you don't like long equations, you can actually simplify that and say that's roughly P molten multiplied by 1 minus P molten And that actually is not even an approximation So that expression should then be equal To this one that we're reading from the plot But the key thing this is what we're measuring per unit And this is what sorry That one we're measuring from the amount of protein we have and this one we're measuring per melting unit And then in theory if you had a computer you could put in everything now, but again, let's do this without computers. We know Exactly in the middle where things are folding if you don't know more than I do Let's assume that 50 percent of the protein is molten there So then P molten is 0.5 the 0.5 multiplied by 1 minus 0.5 Everything is because really simple so And then that means that that derivative is roughly 0.25 multiplied by a term that involves the energy and then kT And then we say that that has to be equal by 1 over delta T And that means that we can calculate how much what is the energy melting energy per molecule And then we also know the melting energy per unit Sorry melting energy per folding unit and melting energy per molecule And then we have those two things and then we can compare them Let's see if I had a no And the point is when we do this For it's not universally true, but it is true to 99 percent of the time that for virtually all proteins The melting unit is if it's a small protein, it's the entire molecule But in other cases it tends to be the domain did lucy talk to you about domains in bioinformatics So for small proteins like christianne anphinsens Luciferase or so if it's a protein up to 100 residues or so there is just one domain For very large proteins say like a ribosome or something that could be 50 or 60 chains So that genes evolve Proteins evolved by genes digging together multiple of these domains and domains are partly the part that evolution evolves So evolution does not work in the her that changing one amino acid at the time that too happens But it's a relatively small component Evolution mostly happens that one gene borrows an entire a part of another gene and takes an entire domain And a good example might be you might have a channel I'd hopefully lucy talked about iron channels. You might have a channel that is ph regulated in a bacterium It's a very simple channel In your nervous system you have almost exactly the same channel But then you have an extra domain so that is voltage gated But the channel is the same. It's just nature has you stole a domain that is sensitive to voltage Many of the things that work in your eyes say the rhodopsin molecules It's the same thing there that you don't evolve function, but you steal function from another gene And whether that's a freak of it's not a freak of nature and it's not a coincidence But the point that these domains are virtually always the folding units too So a large protein with 1,500 residues It does not work. It does not fold by 1,500 amino acids boom collapsing together and folding They work by multiple small pieces on this long jarn or thread They fold independently to first approximation and then these small components for larger agates And the reason we've been able to find that out is exactly because we were able to compare What is the energy per melting unit and the energy per molecule? The beautiful part of that to me I think it reconciles on the one hand physics and on the other hand say biofumatics and evolution and that's a key concept that Biofumatics does not work because you're using the loss of physics biofumatics works by cheating You you're basically looking in a bit litter you're looking at And biofumatics works because proteins are evolutionary related And it's interesting because today everything with big data is very popular, right? And there are some concepts that surprisingly they actually do work that People are because how successful biofumatics has been in the life scientists scientists are using deep learning To for instance predict properties of material based on the properties of another material And in theory this sounds great. You should just assemble a gigantic database of all the materials And then you should be able to predict all the properties But the problem is that silver is not evolutionary related to gold, right? So in some cases it works But biology has this very peculiar thing that all genes are related to that it might be so far away And it might be so distantly related that you can't say it but in principle we are all related While on the other hand everything we've gone through in this course is based on physics that the fact that What where the proteins fold is determined by pure loss of physics and They're kind of two sides of the same coin, right that the physics is certainly true While the biofumatics is frequently a more efficient way to get to the answer But we need the physics to explain why it happens So if we the other part that I that Lucy might have skipped over a bit is to look a little bit Why do proteins unfold or melt or denaturate there are many words? And the only way to understand this is again Go back to this and you've probably heard it a few times now But this is the one equation that you can't forget e minus ts Any particular Don't forget the deltas. So there is a before and there is an after So if you denaturate well the order of the arrow here is important Denaturate means that we go from folded to unfolded And one way to do that is frequently to increase temperature or so And this is partly based on the hydrophobic effect. This delta e will increase with temperature go up Uh while delta s will be positive when we unfold. It's better. The protein will be more disordered when we're unfolding it, right? so that makes sense to understand it but Then there is this complication that s will also drop with temperature So if you just look at that equation what we expect is that at some point If you boil a protein they will certainly unfold, but at some point if the temperature becomes low enough The s part Should start to go in the opposite direction So with this equation is basically saying that if you just drop the temperature far enough proteins should also unfold at very low temperature So there should be a temperature narrow regime Is that true? It happens all the time right when you go out and it's freezing. Well, not this winter that and your proteins will Denaturate and we take an egg and you freeze it It ends up being boiled when you freeze the egg Well, you can certainly get cold damage, but cold damage is usually slightly different that the cold damage occurs when When the water in cells freeze and that tends to explode the cell walls or something. So that's a slightly different property uh But the cool thing is that there is cold denaturation of proteins. So when I was roughly your age This was super cutting it. So it was just the first few cases where people had been able to identify it So basically you need to separate all the polar versus non-polar and drag to forget about the top on the left right But the point is that there are cases where we've been able to see cold denaturation in experiments Uh, and the way you do it is primarily you try to measure the stabilization energy And in many cases you actually you don't you might not see that it becomes so bad that it's going all the way To actually unfolding, but you can see that the stabilization energy of the protein is smaller and smaller and smaller And just by extrapolating you can start to say that well if this were to continue Actually, this might be a better way of showing it So all these proteins at zero they have not unfolded yet, but if you just see so here they would Denaturate due to high temperature between 65 to 90 degrees This regime they're stable in But just based on myoglobin for instance the green curve you can probably see where that is going, right? So just based on those curves you would expect that at some point here they should become unstable again at minus 34 degrees So why didn't they just continue that curve to the left? That's a bit stupid. They should have done that when they published the paper Yes Yes proteins need to be in water and what happens at zero degrees centigrade, right? The water freezes and absolutely nothing happens because then the water molecules stop moving And at that point the protein will not denaturate anymore So the reason why we're not really seeing this is that something else you have a phase transition in water actual phase transition in water first So that's why it's not really that important On the other hand there are cases where water is colder than zero degrees. For instance, if you have a high salt concentration, right? So in the water say and outside Antarctica or something you can certainly have water That's minus five to minus well minus five six degrees centigrade Simply because you have salt So what if you're a fish living in the Antarctic? So it turns out that quite a few organisms have what you call cold shock proteins So these are protein structures that have evolved specifically They're similar to other but they have evolved to be stable at low temperatures and in some cases If there is a particularly cold environment The organisms will even start to express more of these proteins to pretty much protect their cells and everything It's essentially an antifreeze protein So they have evolved different amino acid patterns that are simply stable at lower temperature than in your cells It also depends a bit of an animal right that you're warm-blooded So your blood is always around 100 degrees Fahrenheit But if you're a fish swimming in water, there is no way you can maintain the blood to be at 100 degrees Fahrenheit So that's organism will have to adapt so that the blood can be at minus five degrees centigrade I Think that's all I had from last week and I'm going to continue to the next part which is really fun Because now we're going to start applying everything we learned. There's a bunch of physics here There's going to be a bunch of horrible trial and error and sweeping things under the rug because now we need to make things work in practice But this has become super interesting with the coronavirus so Drug design is a very large process and in particular lots of this occurs in the hospital and everything And I won't go into too much detail there, but I'm going to cover a little bit how we use all the tools that you have learned in drug design In general, you're going to need to start with the bioinformatics part and I'll come back to this at the end But for now I'm going to skip over the bioinformatics just a little bit So assuming that you have some sort of model of the protein that we would like to target Then you can definitely use all the physics based tools that you have To first build all the side chains and everything you can energy minimize it You can simulate these models and at some point you would like to use free energy methods to calculate What is binding here and can you get something to bind to change how this protein works? And that's pretty much how all drugs work. You need something small to bind to change how the protein works We're a bit simple that way. We haven't gotten much further But what we would like you've probably all seen how slow simulations are in many cases You're going to be you have no idea how many drugs there are The space the chemical space is roughly 10 to the power of 60 or so There's virtually an infinite amount of potential drugs So we can't test each possible drug in a simulation So we're going to need to find some sort of sloppy way to rapidly test things that might be interesting And this is where we're sweeping things under the rug, but our rug is a large one So we have lots of room to sweep things under What these drugs will in general do that there is a sort of target protein that's red here And that's there will be some sort of pocket exactly how this looks will depend And then there is a drug that we hope binds and when this drug binds there should be a biological response And this biological response could also mean that the red protein no longer say transport science or something something should happen And one of the reason why these pockets are much more common than we think that this is also how nature Regulates proteins and signaling in general things bind so it's the pockets are already there We just need to find new ways to interact with them So here are some examples up on the left. You see a Pocket in a globular protein. So it's a very deep pocket And it's also hydrophobic and there you have a blue molecule that binds virtually perfectly This is a bit more complicated pocket just on the surface of a protein where you have a small Molecule here binding and this I certainly wouldn't have guessed it, but the computer is smarter than I am And in general, you might have very large Receptors or something on the surface of cell and then it's just a matter of finding what is the exact property of your binding on the surface Your body uses this all the time to your immune system an antibody An antibody recognizes certain antigens and an antigen is just a fancy way Save when you're getting immunized for the flu The immunization just consists of we are injecting small pieces of the virus in your in your body And then your body will express antibodies to identify those pieces And then those pieces will be essentially be transported out But now you have antibodies in your cells that recognize the Parts of the protein related to the influenza virus And when you actually get the influenza virus now, you already have antibodies and that means that your immune system can fight it But this is our way of trying to invoke the same response You would imagine that scientists have been really creative here Right and we've identified things that hit virtually every single protein in your body all types of diseases And then you're incorrect because we're not really that creative The entire blue quadrant up there is deep protein coupled receptors The one quarter of all the drugs known Actually one of all the drugs that we sell just target deep protein coupled receptors So that if you if you ever get a if you ever have to make a bet what the drug hits Say deep protein coupled receptor and it's like 27 percent chance that you're going to be right The these are nuclear receptors and then we have ligand gated iron channels and voltage gated iron channels So there's pretty much four classes of membrane proteins corresponds to 50 percent of all known drugs Not sure these aren't our classes of proteins There's more than one protein there But and then you see and then there's a very long tail of very small components some of the cancer targets are up there If we could get better at identifying new classes of structures and finding new things mine There's a remark or there's a gold mine here of new things that we could use to fight disease It's not at all that these are the only ones that work These were just the ones that we got started with There are a couple of things they're going to need to know and these are typical things that people nasty teachers love to Ask about at exams. There are a few ways drugs can interact with the molecule So by far A small drug we don't there may we call these drugs many things a drug is really something that is sold On the market and particularly used for treatments So we actually give it to patients and I frequently we frequently say drug design and everything That's because we are trying to design something that could become a drug in the future But we need other names for things where that just bind so Any small chemical molecule we frequently call them compound just to make sure that just to tell Well, technically a protein is also a molecule, right? But to stress that we're talking about the small molecules here. That's what we occasionally call them compound But then we like names So any occasionally we call them ligand too Ligand is something small that binds to something large and starts a response or something So compound or ligand or drug. It's really the same thing And then I'm going to introduce a fourth name. Sorry about this If this was me, I would have used one name, but I didn't use it So in particular in the case that we have a receptor and you have a small ligand that should bind and create a response And this could for instance be to create a nerve signal or something these ligands are typically called agonists And an agonist could be say a signal should bind to a G protein coupled receptor and starts a Well signaling in the protein or something A full agonist is something that activates the protein and it activates it to 100 percent As you would expect that it should the normal agonist in your body. That's typically a normal agonist You could Imagine designing an agonist. There are cases that I might want to Activate the receptor the same way your body does So maybe I can design another small molecule that has the same property My molecule will also be an agonist So it will also it will turn the key and start the receptor and it would start it fully Occasionally though, I only get a partial agonist a partial think of a partial agonist that it starts the car, but only to 50 percent Sometimes that's good because sometimes I just sometimes I might not want the full response. I just want to evoke a little bit of the response Or it might be that It's not that good yet I would like to have a full agonist But I I haven't been able to design one yet, but it gets you part of the signal You can have a neutral you can have something called an antagonist an antagonist is really It's fairly simple. This is a binding site. So think about taking a piece of chewing gum and stocking putting it in the lock That will prevent you from putting a key there, right? So you can't open the door anymore And an antagonist is just a small molecule. It binds in the same place, but it does not activate the protein But now my molecule is bounder. So when your actual correct molecule comes You can't bind there because there is chewing gum in the lock If I want to turn off a receptor, that's what you want So an and a whole lot of drugs in the world are antagonists and we Okay, so if we call that that the activate when we start the the The process or the receptor we typically talk about this as inhibit it So we're inhibiting the response. So there is no longer any response So one of the famous drugs in sweden is low sec pre low sec, right? And that's a proton pump inhibitor so that it prevents there's some pumping protons in your stomach And occasionally this is rare, but it can happen that in some cases you might actually want to create the opposite response And that's that's what you call an inverse agonist. So an inverse agonist It has an effect, but it has the opposite effect of the natural ligand So understanding this differences between an agonist and neutral antagonists And an inverse agonist is important to be able to understand what people are speaking about when it comes to drug design And at this point you might think that it's easy. We just need to design these small molecules I'm going to come back to this in a second, but it turns out to be a nightmare in practice that We definitely need a small compound or ligand or agonist whatever we call it that should bind to the target protein And that is by far the easiest part that we're going to talk about today, but if you're Working in a pharma company. The problem is not enough to bind to the protein Any idiot can define something and trust me, I've done it Any idiot can define something that binds You also need to make sure that it doesn't bind in other places Because if somebody were to go on and put chewing gum in every single lock of the building the janitor might get a bit irritated The other part it's say People tend to argue that it's an advantage which actually can get the compound to where it should be say the brain That's not as easy as it sounds We all like to eat drugs, right? But you have a blood brain barrier The blood brain barrier source is to make sure that things don't go from your blood to the brain Your entire stomach is based on denaturating proteins. It's pretty good at denaturating other molecules, too There is an easy way around this you can inject But again, if it's if it's something that say whatever that if you have Assuming that you wake up Monday morning with a headache You don't want to have an injection because you have a bit of a headache, right? So that you want compounds that you can take orally You would like this to be easier to get into the body you would normally prefer not to have to eat 10 kilos on them per day It's big nothing else because a bit heavy to carry You want the slow instead of release of the drug and then it's a huge advantage if you're not vomiting due to the side effects And again, I'm not joking here that if you have a serious form of cancer, sorry You might have to accept those as a side effect of chemotherapy But again, if you have a headache, you're not going to be happy vomiting while you get rid of the headache So in practice, what's it called was it called ad metox absorption distribution metabolism excretion and toxicity? This is actually frequented the largest bottlenecks for drug design and you need to be aware of that But we're not going to go into detail about it. Yes I don't know. I literally don't know Now, of course, sure the body your body can produce 60 billion types of compounds, right? That would be complicated. So sure the body tends to reuse compounds and that's definitely part of it I guess another part has to do with binding that our proteins have not evolved to bind drugs because drugs were not part of evolution So that the things that tend to bind they tend to be small and hydrophobic compounds They tend to be fairly rigid and there is a reason for that if they are not very rigid If it's a long chain then they would have a very good entropy if they are not bound So to be able to bind they need to be fairly small and rigid so they don't lose too much entropy when they bind And simply chemicals the the not the fraction of small molecules that fulfill this they tend to bind in very small places But that's probably the best explanation I can give It's a bit irritating So there are And if we're really going to go into hand-waving territory, there is something called Lipinski's rule of five and the reason why they're five that it's not that there are five parts to the rule but In practice things that tend to make good drugs is that they should weigh less than 500 dolton So they should be small enough to be transported everywhere in the blood Long p should be less than five and that has to do with the partition coefficient with octanol and water And that means that if it's a very hydrophobic drug you're not A very hydrophobic drug might be great at binding But that's not going to help you if it can't be solvated in the blood then actually get there in the first place, right? You should have less than five hydrogen bonded donors Don't ask it's just Yeah, that that usually works Uh, and then we had to well, it's almost five but it's 10 So we multiplied by two less than 10 hydrogen bond acceptors And that both of these hydrogen bond donors and acceptors well The rest of the drug should be reasonably nonpolar so that it can also cross membranes So it can't be too hydrophobic and it can't be too hydrophilic and then it should be small and rigid Historically, this has kind of worked. The only problem is that it hasn't really developed a new drug in 20 years There are many reasons for this one of the reasons that we are much tough for regulatory requirements on drugs today Aspirin would never have been approved today. It's far too dangerous It's super dangerous because you can lead to bleeding and other things But drugs that have been approved historically we don't unapprove them So the problem is that we're getting higher and higher requirement that we can't have side effects Everything needs to be tested and everything. There are fewer and fewer drugs It gets very expensive to develop new drugs and of course we start by picking the lower hanging fruit, right? That these you once we've already found So having said that there are These are examples of for drugs and you've probably never heard about these names But so that the chemical names There is a chemical name and then there's a brand name. So low sec for instance That's called homeoprasol. Homeoprasol is the actual chemical name But depending on the market you're selling it. It's called used to be called nexium in the us While it's called low sec and that's that's just pr and whatever brand they can protect And here you see that the uc's effect that they are small. They are typically fairly rigid with lots of Rings ring structures will be rigid and then you can see that many of them target Various receptors here and that a couple of different Indications so the indication is word to use for what is it? What is the reason for giving administering this drug? So, how did they find these drugs? Where do they come from? In principle that could work. The problem that chemical space is very large, right? And as I was about to say divine inspiration is fine and literally it is fine if you if you come up with the drug Tomorrow that would treat the corona virus. I think people would be fairly happy They're not going to start to ask you. Well, you need to be able to prove how you came up with it If it cures the corona virus, it cures the corona virus. We're fairly happy if it doesn't have side effects The problem is a divine inspiration when you have 10 to the power of 60 alternatives that you can't we can't search randomly Historically this used to be fairly easy that we used to find things in nature Quite a few drugs have been found for instance the amazons and everything that there might have been a tribe that Well cocaine is an example of this right that you're finding a substance that you're chewing leaves And that means that you get rid of the drowsiness And then eventually you purify that and in that case it actually became a different type of drug And what that usually then evolves into that you need to identify what is the active component Because you start out with a leaf or something that has some sort of component We don't know which one and then you go through a process of isolating What is the specific molecule in this specific plant that has a component? But by the time, you know that there is something in this leaf. It's fairly trivial to isolate what it is Then in many cases these components are not particularly efficient But again, then you have something that works and then you follow exactly the process that you said can we be smart Can we optimize this a little bit and make it better? And occasionally you can if you know where it binds and you know that oh There's a hydrogen bond donor here, but if I put the hydrogen bond acceptor on my molecule here It would bind even better and then if you're lucky it does So there's still a bit divine inspiration, but Slightly more guided The problem though is that here too this was easy for the low-hanging fruit But as we there are relatively few tribes that have identified plants that are good at curing say Cervical cancer or something right that as we're getting into more advanced treatments There are fewer things in nature that will work So If you look at some sort of modern drug design We don't just go out and look at things serendipitously because in this case for modern drug design There might be a specific disease for instance coronavirus that we would like to go after And then we need preclinical first. We need a bit of semi divine inspiration We need something to start with and I'll come back to how we find this And that's hopefully we have something that has at least a little bit of a clue this might have a chance to bind And then we need to go through and iterate this and try to improve. Can I get it to bind better? And the neat thing here is of course this we can we can do some of those things in the computer While other things we do in the lab if you have a great idea for 10 drugs that you think might work I can go and test those 10 in the lab. I can't test 10 to the power of 3 in the lab, but I can test 10 And then you have to do first tests in chemistry and eventually animal tests And this is a process. This is actually fairly cheap and this happened predominantly happens in academia today At but at some point here you realize you're super excited and this is where people start to say they have developed a new drug for whatever Curing cancer and that's when you see universe that is putting this on their home page saying that it's amazing Group X has developed something to cure Y It's nowhere near a real drug yet. This is a very early candidate So what then you need to go through clinical testing and the phase one you need to says is this safe for humans to take? And if humans die from your drug, it's likely not going to be a blockbuster Even if it is safe for humans literally here. We're not even worried about whether it cures anything We just want to make sure that people don't die or become seriously ill from taking this If that is the case the second that does it have any effect whatsoever on the disease in humans There is a famous twitter account that says The title of the account is literally just in mice Because again, that's what they heard that there is this The classical thing is that the university will pay that Whatever eating low eating a high calorie diet will make sure that your sport performance improves And then there's an illustration of somebody running But then you actually read the paper. Yes, this was determined in mice And I hate to break it to you, but you are not mice So that quite a few of these things it works in mice models, but not in humans And there is another joke that some colleagues thought that If you're a mouse and get cancer, we have very good treatment for you Because that's where we've optimized everything for it But even this is not enough The question is it's fine if it works, but is it better than something we already have on the market? Because sorry coming in number two and asking to charge 10 times as much for something that is not as good as what's already on the market You're not going to get it approved. You won't be allowed to sell it And the point is these things are super expensive and this entire process can take a decade And if you were occasionally worried about your performance in classes and everything Farm trust me big farm is far worse. The red part here is the fraction that failed So in the preclinical states like roughly two-thirds of everything failed During in-vitro tests and test tubes and animals. This is awesome Normally, you would think that wouldn't you prefer that to be green on the contrary failing here means like you might have a very disappointed student Or advisor, but failing here might cost you ten thousand dollars Failing here cost you a billion dollars So if you're a pharma company, you want to fail there failing early is failing cheap The further you go here The catastrophe is that when you're 500 people working on the project you bought the company developing the early drug You've taken 10 years and you bet the entire market on this. That's where you don't want it to come back and say sorry It was not as good as something that's already on the market That's when companies go bankrupt and you see this. Okay, even in sweden You see this occasionally in the stock market where there's a stock that drops 60 percent Because suddenly they came back from a phase two or phase three test and that was negative So that's a real nightmare The typical examples that it might take 10 to 15 years The problem here is that You only have a patent for 20 years In the western world and you need the patent things before it's known so that once you start selling the drug You only have five years to make back all your costs This has been extended a bit so that you can I think you have a five or eight years extension Precisely due to this story, but this is what sorry. This is why we have to pay as much for drugs as we do 300 million euros doesn't well I was about to say that it doesn't sound so bad. It sounds really bad But this is the cost of developing a drug Then there are plenty of drugs that fail to So if you're a large company if you're successful in one case out of three, you're probably very happy And that One case out of three that is successful will have to recover the cost for the other two cases that were not successful To and all the cases that failed really early so that Maybe 150 scientists are so involved. So it's a fairly large very expensive operation and the second this comes to market You only have five ten years to bring back all the costs Yes I'll come back to that There are It's complicated and we start to relax some of the rules, but trust me, you don't want to relax them too much I think I have a few slides about it where things can go horribly wrong Uh So the research phases that we've got through that we're primarily very early on in the discovery phase or something Then you might be looking at genomes. You look at target validation. You might try to determine structures of them We're going to talk a little bit about high throughput screening trying to identify things that bind And then eventually we come to leads and then at some point you start going into the clinic I won't really cover the clinic here anymore But the point the earlier you are the more we are moving towards computers So 10 years ago, everything was experimental 15 years ago at least Today, I would say virtually 90 of the really early phase stuff is computers because it's cheap and we can do it at a much higher throughput than we've ever done before Um I think this is a good place to take a break. So after the break I'm going to go through first what we do in these different Faces, uh, we're going to go through some of the challenges how we can fail I will I think I have a few slides of it But if we don't I'm going to deliberately going to bring up some of the things to talk about when things can go It sounds really great that we should be better at imminent threats That can go horribly wrong in some cases and then I would talk about coronavirus too because there's a ton of cool things happening there Let's meet here at a quarter past I realized those side effects. I'm going to cover in a lecture tomorrow. So you will have to wait 24 hours for that But to go through this I'm going to start by talking a little bit about the computational tools and the protein structure parts we can need here So at some point we need this could literally be divine inspiration But we need some sort of molecules to start from And they don't really have to be that good, but it has to be better than nothing. There has to be at least a trace of an effect somewhere And historically this would be that's something we found in nature And then we try to get rid of side effects and get it better Today we typically do what you call high throughput screening and this is an experimental procedure So these really fancy robots So that if you have some sort of anything that you can measure and this could even be measuring current through a channel or something Then these robots can run maybe 100 000 tests a day. So they test 96 wells at a time Very high throughput. It's They are not cheap Actually, they are cheap per well The problem is that you run through quite a few wells per day, right? So that Operating these machines can come at the cost of a million and a half Swedish corner per day So, yes, you will have them if you are a former company But the cost run up and they run up fairly quickly and the problem is 150 000 is nothing compared to the 10 to the power of 60 So that we have to be smart about what we're testing The other point is that we don't you can't Testing it here is cheap, but we don't have 10 to the power of 60 molecules So we want to start from molecules that already exist and there are a couple of compound databases One is called zinc for stands for zinc is not commercial And that has maybe a few hundred thousand molecules or so that we can order relatively cheaply And if you're lucky from this as a minute, it's not that hard to find something that finds if you're lucky you get 100 leads here They might not be great, but it's something to work with But the problem is that it costs quite a bit, but historically this is what always starts with And in principle give it compared to the Compared to the volume of chemical space It's a laughable probability that we would find something good and in some cases this is an example that a colleague in Uppsala Extracted a few years ago. They were trying to find something that's bound to lactamase They spent in the ballpark of half a million dollars And they came up with zero hits That's not what you want to go into your manager and say that oh by the way joe That's that last round that causes half a million dollars. We don't have anything for it And then this second case you were lucky hundred 46 it's for good sign The question is can we do this better in a computer? And i'm gonna live away. No, we can't but we can do it way cheaper in a computer and those things are a bit coupled so What if we could have there are a couple of tricks here that maybe we can just find some sort of regression if molecule a works Molecule b might also work and that's relate something called Pharmacophore and then we're going to talk about not high throughput screening but virtual screening So the difference between experimental and computational space is that The computational parts can be super fast We in an experiment where we might be able to do 300 000 compounds In the computer might be able to do a billion compounds Now the experiment is much more accurate. There is no question about it But it doesn't help you if you're accurate if I can screen 10 000 times more compounds than you do right Because if we don't have to be super accurate I just need something to start with and being able to test 10 000 times more things will give you much greater chance of finding something So the idea is not to go for accuracy, but do is sloppy and fast And there are some simple ways of doing this so that An obvious way is anesthetics. We know that good anesthetics tend to be small and fairly hydrophobic So if I ask you to find me other molecules that might work as anesthetics, what type of molecules would you look for? Once that are small and hydrophobic, right? And there are some fancy words for this that you can do this more advanced, of course And this called something called QSAR quantitative structure activity relationships And that's pretty much just a relation between the expected biological activity to simple chemical properties Hydrophobicity the number of hydrogen bonds maybe the charge the molecular weight Does it have an aromatic ring? And you can encode all these things in long tables so that a computer can understand it But it's the first experiment is really just fancy linear regression And a few years ago that had fallen out of grace a bit because linear regression that's for Linear regression is for cc's that no real scientist does linear regression But then you've all heard about deep learning artificial intelligence and deep learning is really it's not linear regression But it's very advanced regression. So this is exploding again now So what what everyone does now if you have 10,000 molecules that you have measured something for People try to use deep learning. So based on that can we have the computer predict what will work? And it probably works slightly better than QSAR, but it's not a miracle Oh, sorry. I had a few examples there, but the whole point simple things But you can do it slightly better if I know my target if I know the iron channel I might I might be able to create a three dimensional map I might be able to say that look here. I really need something that's hydrophobic hydrophobic hydrophobic Here I need a hydrogen bond donor there. I need a hydrogen bond acceptor And then you can kind of map this out between distances between them in space That's a slightly more complicated. We call the a pharmacophore But both of what these both have in common is that they're not trying to solve all the physical problems They don't go the door in details They just try to first approximation roughly what type of molecule might fit here The problem with QSAR though in particular is that advantage is that it's super fast the disadvantage is that If you make a very specific model here, you will only discover things that are exactly like what you already know And in the case of anesthetics, some of the best anesthetics today are actually not purely hydrophobic ones And of course, if you only looked for hydrophobic compounds, you would not find any ones that are not hydrophobic Large molecules that are flexible, they might occur in different Conformations and everything and that's not really included in this model So what if could bind to multiple sites? So QSAR kind of fell out of grace because it was not really advanced enough So where QSAR only describes the average properties of molecules these three dimensional patterns If you encode that in tables, we end up with something called the pharmacophore And a pharmacophore essentially just say that you need a hydrogen bond say 10 angstroms away from a aromatic ring that should be five angstroms away from another aromatic ring That should be 10 angstroms away from an oxygen and now I was just inventing things there right But some sort of very simple way that you can encode in tables This does not describe everything about the molecule mind you, but it encodes some of the three dimensional properties of the molecule And once we have this we can start screening through those gigantic databases What are all the potential molecules that you could imagine that we produce would fit this And sometimes this works You might also be able to find that there are common elements. I already mentioned to you that It's very common that you have small and fairly rigid compounds. So all these aromatic rings keep coming back So this is an example of a whole range of it's a series of drugs that are kind of similar And I think you could encode not everything here, but many of the things here would have similar pharmacophores And exactly how you encode this varies a bit, but it's a very simple way to try to describe overall three dimensional properties We might also want to say roughly how much space does this molecule take so that it will fill out the entire binding pocket And this far I cheated I haven't spoken anything about protein structure because this is so dirt and simple that you completely ignore the protein If I have some drug then I can try to find other things that look like it And that's very much based on what you asked before that you start from something that's known But what if you don't have anything that's known? well In some cases you can start from the protein and if we Have a structure of the protein we could do what we call molecular docking So say that we have a dopamine receptor or something that if we have something that binds here We could in theory put a molecule here and rather than go the pharmacophore route try to identify the binding energy here Even with free energy calculation, but free energy calculation would be too expensive 10 simulation is not going to cut it 10 000 will not cut it. We need to be able to run a billion or 10 billion And this is fairly easy. You've all done this In a slightly insert the part on the left. We've all done And this is pretty much the same thing but in the computer That we literally this has to be so fast that it takes one second Promote you we want to test and then we test we have a super computer And we test 10 000 of them in parallel and we try this like a million times that gives us 10 billion So just take a molecule and try does it fit here does it's kind of like your average six months old Right that you try you try the square one in the round peg and then you try to push harder But it doesn't work and eventually you try it in the square hole and then it works This is exactly the same thing take the molecule does it fit there does it fit there does it fit there? Does it fit there? No try the next molecule and when you repeat this 10 billion times Eventually you're successful and find a square peg in the square hole maybe But to do this you need a structure of the protein at least an homologer model But ideally a crystal structure and that's why quite a few pharmaceutical companies They spend millions if not billions of dollars to determine structures of important targets And docking in a way it's very simple that what we need we need the best way to put two molecules together So we need the best way and we need some way to saying how good it is We typically don't use a formal force field here because that would be too expensive So I need a very cheap way to say was this good or bad And then I need a way to ways to put two molecules together that If they can be in multiple orientations or something you need some sort of way of sampling and testing things So we need to search So we need a very simple fast way to say whether a particular test is good or bad and I need a very fast way of searching And in principle your guess is almost as good as mine here, but you need An obvious way to sampling is that you might take a small molecule and just try to randomly rotate it So you have the entire molecule first and then we might have a few Bonds inside the molecule that can rotate and then I need to try to rotate those bonds too And again if I'm lucky I might be able to do this so that it takes a second or a few seconds per molecule And divine inspiration is perfectly fine here. I don't care if I miss something good that happens Because like if you again if you look at lousek You don't need to find the world's best molecule to inhibit the proton pump All you need to find is a good one So of course I would it's a bummer if I miss something really good I would prefer that I don't do it, but it's more important that I find something And that's why speed is everything in docking so that all the beautiful things in physics we throw out the window here It just has to be super fast Uh, but if we were to do this exhaustively for every single patch on the surface and it would take 200 years We can't do that So then you end up with some sort of algorithm that we occasionally call them genetic algorithms But the only part that's genetic here is basically we try to have the algorithm learn So that we make an initial population randomly That we might try to place it randomly and then we over the surface and then we realize look In these two or three patches we tend to have fairly good energies and that means focus your attention there Same thing with the molecules if you test Tons of different classes of molecules, but we realize in general the hydrophobic ones tend to do better Focus your attention on the hydrophobic ones So we try to have some sort of self-learning step just as evolution of self-learning here, right? And again, this will mean that we miss some things But it's better to find something than waiting 200 years before we get the best possible answer And then we just Well, whether we call it mutate, but we have to sort of change some things see me randomly and then repeat repeat repeat There are other ways to do this that you don't necessarily need to start from a full molecule You can do what you call fragment based drug design so that I can if I have a small library of multiple fragments I can try to start from my pocket and then I try to build different fragments in different parts of the pocket And then when I find a bunch of good fragments Then we ask an organic chemist. So could you stitch these fragments together to one molecule? And in this case at first dates the organic chemist is a computer too So we ask the computer can you stitch this together into one molecule? So then we gradually build up the molecule in the binding site and that also frequently works well And then to combine that we need some sort of ways to very rapidly say which one is good about You could use one of these MD force fields, but that's too slow. We don't have water here So in general we tend to have something that is very ad hoc We say that hydrogen bonds are good if it's hydrophobic atoms matching hydrophobic atoms, that's good If it's a positive charge matching a negative charge, that's good and vice versa when things are bad And there are few there are a few different forms of doing this You can also make some sort of statistical potentials based on how likely is it for pairs of atoms to be close to each other And we're well aware that this is sloppy, but it has to be fast fast fast And then to make this even faster, we don't try every single position, but actually might be easy to show you this way We might take some sort of molecule I might evaluate properties on each grid cell saying that that particular point for this protein tends to be hydrophobic That is hydrophilic hydrophilic hydrophobic hydrophobic charts And if you do this, then I can just take my small molecule and test this against the grid It's horribly sloppy. It's that's the things that would make you tear your hair, right? But it's fast fast fast and if we find something that's slightly good Then we can decide to take a step back and do it much more accurate So it's you only you can do anything you want, but it can take more than one second per molecule you can test And the point is that This kind of works. So this is the same example used before Lactamase you spend three you spent half a million dollars zero experimentalists, but they found two docking heads The computer time here probably cost you one thousand dollars And I'm not sure about you, but if I were the chief financial officer in this company, half a million dollars versus one thousand dollars A thousand dollars is a hell of a lot nicer Here the docking was not at all as good as the experimental one On the other we don't know how good what if the best possible hit was one of those five then we're happy, right? That we don't necessarily need 146 So what we typically do here is that you start out in the computer Based on then we take those five hits into the lab and see are any of these interesting If two of those are interesting then okay Let's come up with more molecules that were similar to the two interesting ones and then repeat this again in the computer So that this the point is not that the computer is replacing the lab But you iterate between the computer and the lab and use the lab to check where the computational prediction is good And as you get slightly further you can allow even the protein to be flexible so that the side chains of the proteins can move The cost of that well the advantage is that you will get a better prediction The problem is that now it will be 10 to 100 times more expensive So suddenly you can't you might be able to test 100 million compounds rather than 10 billion So this is a trade-off that you would like the accuracy But it's more important to find something. Yes, I mean try the experimental hits in the computation Yes, so normally we start in the computer because it's a hundred times more cheaper, right And and the idea normally you would not like five you might testing 50 or so works fine And then we want to try those 50 based that most pharma companies they work in a flow that takes roughly four to six weeks So that during those four weeks based on where the structure is now You need to come back with the within two weeks and that's like 10 working days Give me a list of 10 things we should test in the next experimental round So that you can do anything you want in the computer lab, but it can take more than 10 days I'm completely uninterested what you could do in three weeks So in 10 days I need to know what we should test in the next round and then we do a new experimental screen It might be 100 compounds based on the outcome of that that you will know another week later Based on the outcome of that you can go back into the computer room and decide What would you like to do now because now I want 10 better things to try And then you keep iterating this way year after year after year to improve it At this point you might have a drug because you actually have something that it binds and it binds and we know that it binds And it has hopefully been confirmed experimentally that it binds to the right receptor. This will work And it might even inhibit the receptor If you're fine with eating five kilos of medicine per day with lots of side effects And not very strong because the problem is not going to be very efficient at binding And that has to do with the equilibrium constant If it's not a very efficient binder and you want this to bind to every single cell in your body You need a very high concentration in your blood say one molar or something Trust me at one molar, it doesn't matter what molecule you're adding You're going to have so many side effects that that would never be approved So we now need to improve the binding to get it to be very very efficient at binding so that you can take a small pill And the way this works is that you go into the computer and iterate this And then you say try to You might create a pharmacophore. You might try to This is a pharmacophore that identifies something that's similar In this case they end up with a symmetric diol And then you find a database it this was the initial design. There's phosphate groups down there This was they extended to make the diol again And then they added urea groups and then they optimized the stereochemistry to bind in the particular receptor step This was the final drug that was then selected for phase one trial And this is a super famous molecule because it's the hiv1 protease inhibitor It was the first drug to actually have an effect in AIDS treatment And it was the first drug that on the preclinical stage was designed computationally So this is a drug. It was not based on anything in the amazonas or anything It was designed from scratch to target one receptor with computers This is another cool drug Atrovastatin lipeter. Do you know what you're looking at? You're a bit too young to eat lipeter They're basically lowers cholesterol You're looking at 14 billion dollars per year of sales This is probably the largest drug success ever in the world until 2015. It made 140 billion dollars These are the annual sales revenue And then something happened. What happened in 2012? No, no, no that uh high cholesterol they the drug is still it's a great drug They ran out of their patent So at this point it was generics. So then anybody could copy the drug And then people started producing it much cheaper And this is a point here. I'm not Again fighter. They have they certainly made a ton of money here But of course all these absurdly high revenues they paid had to pay for lots of other failures and lots of marketing and everything But you only you have a relatively few years when you're making all the money of the drug And today, of course, they hardly make any money at all from lipeter anymore So drugs have a very short commercial life. Now, of course in the scientific life, it's great. The drug still works So one question is you started use molecular simulation historically We have not used molecular simulation for drug discovery mainly because that's been too expensive That's have been unclear how accurate they were and in particular the free energy calculation methods have not been good enough This is gradually changing the last few years There are some beautiful examples where MD simulations can predict free energy calculations very accurately And I think we are some examples from David Shaw actually. Oh, this yes, this is a brute force sampling of a small drug Binding to a large protein and you will see that eventually this one will sneak into the binding pocket there in the middle Yes, I think there you have it and the cool thing is that when they determined the x-ray structure of this The ligand was exactly in that place And this is by no means cheap. They're they're designing this hardware and they're selling them for millions of dollars But some of the first gpcr experimental structures that were determined are rumored to have been investments So commercial investments of two billion dollars to determine one structure And compared to two billion dollars any computing time you can any computer you can imagine is going to be insanely cheap Compared to taking 10 years and two billion dollars We can learn quite a lot the other thing that simulations can do they can occasionally tell you exactly First, where's the ligand binding? What is the interaction energy? Occasionally can tell you how the activation process happens to but you see here The orange is the one predicting the simulation the black is the one determining the extra crystal structure It's scarily good So the computer we are able to do chemistry entirely in computers today Without determining the co-crystall structure The other thing that the simulations can tell you that you've kind of looked that that's already that Is home cases that how long does it take for this ligand to bind? Is there does the process have to go under through some sort of activation historically? It's not something we've looked at but in some cases not just important that it binds But the question what is it doing once it has bound? How is it changing the receptors role and what is the time constants here? So this is until five years ago I would say that this is something that could be interesting for the future. That's no longer the case This is increasingly used at least to understand flexibility and where things bind the drug design And that brings us to the last 10 or 12 slides here that there is some I'm not sure where the cool is the right word But we're fortunate Fascinating to be the part where we have all this corona virus stuff because this is happening as we speak right now It's a remarkable insight This is completely different from how traditional drug design has worked. I think it's a sign for the future I have no idea whether it's going to be successful or not But this is going in order of magnitude faster than I've ever seen anything in drug design before And that's what makes it a bit exciting and it's also much more computational So previous in the class last time I booked up I call this NCOV 2019 the name was formally changed to 2019 COVID So this is the corona virus that appeared around the turn of the year in China You can actually have a look at some of these in04 which is an austrian company evolving drug design They are updating this website pretty much day by day So you can track it in real time how things are going so that I've stolen some of the slides from their their page And the first thing you do if you don't know anything a virus genome is fairly short I had the corona virus genome on a slide earlier in the class And the first thing you do is that you put that in the database and use bioinformatics tools and identify Do we know anything about this? And in some cases you can find sequence patterns and in this case we were even happier Because it turns out that some of the proteins encoded for by this virus We actually know the structure of So this is a protease Kind of like the HIV-1 protease And the cool thing is we have some like 98% sequence identity to a protease with known structure And if I recall correctly, that's from the MERS virus, which is not entirely surprising. That's another type of corona virus If you have 98% sequence identity, is that good or bad? At 98% sequence identity, I would eat my left shoe if the structures are not the same Like it's so good that you may we know what this protein structure looks like the computer can get this you don't need to determine it So it's great. We know what at least this is not the only protein in the uh in the corona virus But it's a very important one and again in HIV for instance, we have drugs against HIV-1 protease So this might be an interesting target if we could destroy the function of the virus's protease That's a potential drug that we could use So just based on this sequence in an hour you can create an homology model of the protein and they they come through that They've done some molecular models. They've even did some simulations and energy minimization, which I'm not sure how important it was Uh But based on this you can also you can learn a bit about the structure and in this particular case You said the the confidence in the structure prediction is 100% So that 300 use covered there are 2% of my sequence that are not in the previous pdb structure And that's fine that it's either. It's a very small loop Or it's the very first or last parts of the sequence. Don't worry about those 2% This is going to be awesome and you can even bring that if you go to the web page I think you can bring that up and click on it and rotate it to see what it looks like You can also even when you've let the computer optimize this just to show you When you create the homology model how similar it is between these two viruses now mind you this is a model But let's see here I never remember which one is which but it doesn't matter But so that the actual pdb file might be the green and this model is the blue one But the point is that they are so similar that they are virtually identical There are a handful of side chains that might vary a little bit The only problem though is that if those side chains are in the parts in the binding pockets and everything That will change the things they bind and if either one amino acid is sufficient to make the virus Make the protein behave in a different way. Yes, you have a question or Okay, sorry, so somebody waving So the next thing we need to do is we need to identify what are the pockets here In theory, you could start to duck away right away and try to find something But that's a bit stupid that it's unlikely that you will bind here If there is an important pocket here that's important for the protein's role It's likely a bit buried and everything like the pockets we saw So an obvious way to do this is that let's just try to find holes in the protein first And there are some fairly easy ways to identify cavities So they found one cavity down there just with these point clouds and another cavity down there up here In this particular case this company they have their own algorithm that They called what did they call that? I should know It says somewhere It's a catalofor model, which is the service they are selling But it's basically a pharmacophore like model So that for each of these pockets they now have a recipe here So what are the types of structures that might fit in these pockets? And then they're screening through very large databases to find are there potentially interesting things here that could bind in this pocket now For the new model that has all the side chains of the coronavirus structure And at this point this probably took them three four days because you need to create a model model You need to make test a few different models make sure that Make sure that you're happy with the structures make sure that you're happy with the cavities Maybe try out a few different cavities so that you don't make mistakes But it's a relatively fast and not too computationally expensive But at this point now you want to start screening through millions of different compounds So then they used time I think it wasn't a cluster on amazon or something and used 23 years of cpu time in 24 hours So that used a gigantic large external resource and the more time you have the larger a database you can screen through And once they screened through that database they came up with a short list of roughly 10 compounds I think but I think they only showed the first 10 ones So their argument was that anything below 0.1 in their algorithm is a completely arbitrary score where low is good Should be an interesting target So then they had a short list of 10 compounds that could be interesting to try And they haven't again, this was like two weeks ago So they haven't had a chance to try all these things first I'm not even sure whether they all these are available Synthesized because it's one thing to knowing that that particular molecule would be good The question is can you order that molecule on the internet? Otherwise you're going to need to hire an organic chemist and show him the formula and say can you come up with a Way to make this molecule and then the organic chemist will scratch their head and say give me a few days And I'll come back and then they need to invent a new synthesis process to create that particular molecule And depending on how complicated that molecule is that can be a very complicated expensive a difficult process I think I've heard In general if you ask if you outsource the production of a new molecule to ashore So expect to pay between 10 and 50 thousand dollars If in a molecule that has never been synthesized before That's perfectly fine. If there's one molecule you want to test for a coronavirus If you're working in this company though, you likely don't want to go to your boss and tell him I would like to synthesize 1 000 such molecules because that's 50 million dollars just for this week Well, you can if you're lucky they're going to be very happy or you will be out of a job next week But then if everyone tends to rewrite history a little bit so that there were tons of things going on in parallel So that in parallel Chinese scientists that were testing things and I'm it's a bit unclear how they tested it But a molecule that several teams became interested in in parallel was called Lopinavir And Lopinavir was actually part of their top 10 or 20 compounds and it had a score of roughly 0.08 So below below their 0.1 threshold Uh, and Lopinavir is a drug that is actually used clinically to treat hiv This is a huge advantage Why? It's already synthesized and that means you can buy it. That's certainly one advantage There's a greater advantage. I'm not so worried about the 50 000 dollars now Exactly, right? So that certainly there could certainly be side effects of this one But it's something that has been tested. It has gone through phase one phase two and phase three It might not work so that in principle we believe very strongly that we need to stick to science We can't guess And that's important that's super important in general in drug design because that's you might think that that's stupid Why shouldn't we guess sometimes if it's important? I can't say lives that The problem is that hundreds of years of modern medicine the entire success of modern medicine We know what we're doing. We're not doing things randomly We're basing it on science and we're giving a drug because we actually have a clear hypothesis why it's working And sure for one patient you could argue it's no big deal if we just try something, right? But if we start to throw out this whole basing it on science, which we will just try anything Before we know it. Well, you all you might have seen the scandal around matronius, right? That's what happened when we started throwing science out that we will just try anything that might work So Scientists and drug we are very hesitant to try things that are not been proven to work But then there are exceptions and in particular when it's a pandemic or something we might make an exception You will not get an exception to start randomly giving a patient one of the molecules in the previous top five We have no idea whether that will kill the patient and that's deeply unethical Because the likelihood of dying from corona virus is at worst two percent Random molecule that you're injecting it could be a 50 risk of dying so we can't do that But a medicine that we know is safe at least at low doses So this is called repurposing and it's starting to become very important because great We have something that's safe and we know Maybe we could take this medicine and use it for something else too And there's a quite fun fact that when it comes to patents So you might own a patent for lopinavir when it comes to using lopinavir to treat hiv So you're making tons of money for hiv But I if I discover this I might be able to get a patent for using lopinavir for treating this molecule So you are now not allowed to use lopinavir to treat corona virus without paying me Now on the other hand you own the patent to the molecules. I can't produce this molecule without paying you So then we need to sit down and talk maybe for my joint company, right? So that Is this horrible Well in a way it is horrible, but but the point is yeah, sorry Sure, but remember that when it comes to the phase one the phase one testing is that we test it on healthy subjects, right? And that's so you usually take some people who would do anything for money like students Then they give them $50 and they think they're making $50 what you're I'm not kidding You're literally giving students chemicals that I've never been tested on a human before but they have been tested on mice And then the students take $50 and are very happy because they have something to eat that week So that we have tested them on healthy humans And at some point yes, there is still a risk But at some point if we start if we start seeing that this risk is significantly lower than one Smaller than 1 percent with the risk of dying from the virus is 2 percent At some point the benefit outweighs the risk and it might be worth testing this for corona virus The other point that we're not doing this randomly. I'm not randomly giving you any drug, right? This is a virus where we know that this drug Influences and binds to the HIV protease inhibitor. This is also a protease inhibitor It's a very similar protease inhibitor and I now have a hypothesis that this will be able to bind this particular inhibitor too So this is based I'm relaxing the trials a bit, but it's still based on science. It's not random We don't know this word here that there have been some indications that some of the patients are getting better from this The danger here though is that there are so many rumors floating around and of course some patients will get better It's not necessarily it might just be that they got good treatment in general And this is the problem that we need those tests to make sure that statistically that things work In a crisis we will relax it a little bit But long term it's very dangerous to start giving up the test and just shooting for the best and hoping Having said that So what what this inner four company that did they now refocused roughly a week ago on lopinavir So then they used MD simulations Just like the ones you were doing two weeks ago and they found a bunch of different bound conformations in the pocket So these are snapshots of lopinavir bound with slightly different conformations for the lopinavir molecule in this particular pocket I think yes, I managed to Get a small movie from them. So this is a super like two weeks ago or so like three weeks ago There's a super short simulation only 300 picoseconds not much longer than the ones you did And I think you will yes, that's one example of the molecule bound and let's see if I think yes I think there you will see it moving. It's not really moving a lot here They are not using MD simulations to actually calculate free energies or anything But they're somehow using it to realize the molecule might actually be stable there It appears to stay bound there So at least is the first reasonable shot to hope that it might stay put stay put there In a crisis may be good enough Give this a few years in the future I think you would do a proper free energy calculation as berg talked about and say specifically how well does it mine there Today that would be a bit expensive. Give this another five years or so. It will take an hour Because it's as frustrated as we are and as slow as computers sometimes feel computers get roughly twice as fast every year, right? We don't have any near anywhere near that development in experiments Experiments might be twice as good today as they were 10 years ago So and that's why we increasingly seeing this shift. There's more and more and more we can do in computers and it's cheaper all the time In parallel to that there's one more thing The cool thing here that in general we need structures of proteins to so that I we hinted a little bit about cryoem But since since I spoke last week There have been a joint us and chinese team that were able to purify One of the spike proteins, which is another protein in the corona virus in 12 days So that they managed to identify the gene. They were overexpressing the protein. They were purifying the protein They put this on cryoem. They were able to record this. They were able to determine the structure They were able to refine the structure. They were able to write up the manuscript They were able to submit the manuscript to a preprancer in 12 days four days later. This was published in science So the total from the from the point this started the work until this was published in science was 16 days I've never seen anything like it. Like historically this would take two years So if there's a I'm in shock and all at I think this is science to come for science in the future That china is becoming a world power in science. They're insanely good On the one hand on the other part the part of it. These are also us team symbols So it has become a worldwide effort and it's kind of amazing how much everyone is sharing everyone is sharing all the information all the time I have no idea whether we're gonna have new vaccines for the corona virus But it's it's a very interesting development for the future and I think even at silat lab We're talking that in sweden We need to set up some sort of rapid response task forces So when things like this appear we should instantly be able to push the button and have a team of people start working on it So we're not there yet. I think this was published february 19 So we're not there yet But I think give this a few years It's not going to be a matter of a decade to develop new drugs But in some cases we will be able to do it in months Now that brings of course other Talking about all the collaborations is fancy, right? But those 140 billion dollars that fights are made from repeater So that's mostly rich rich fat people in the western world. That's your ideal If you want to design drugs and make money, that's the type of patients you want So make a super expensive drug sell it to people with lots of money And make sure that it's a drug they have to take the rest of their lives Uh, so that you should not the worst drug they have is a drug that you secures the disease Because you will only sell them one dose That but if they have to keep taking this anti-cholesterol thing the rest of their lives, that's awesome um Now of course I'm joking a bit, right that These drug companies are not necessarily god's best children in the sense that they're they're not out there to make a benefit to society And it's very easy to argue that they make too much money. We should tax them and everything Um, I think most of these companies spend more money on pr than they do on basic research They prefer that the university does the basic research And that's typically the sentiment we normally have we like to paint them in a bad picture on the other hand If we compare let's speak AstraZeneca a swede. Well, it's half swedish company Is AstraZeneca a bad company? It's easy to argue that based on how much money they make but if you compare that say to Bufors, they're selling weapons And everything is relative right that compared to and again, these are scientists There are people working with management and everything they're they're professionals that they need to make a salary They need to pay their mortgages on their house In the grand scheme of things making money from curing ill people I find it difficult to fault that compared to say making money from selling weapons And that's where things get a little bit more complicated The reason why this entire process is very expensive is that we don't accept we don't accept side effects We only give them a relatively short patent protection. We want the new drugs We typically don't pay for them ourselves. So that but we just expect the drugs to be there, right that I don't think it's quite as clear cut as you occasionally make it In the news There are certainly some really horrible drug companies out there that just buy by up old medicines And then they stop producing them try to increase the prices, but those are the exceptions rather than the norm So some of these companies invest quite a lot in Not only do they invest quite a lot, but they're actually able to cure diseases that we couldn't cure a few decades ago Leukemia is a great example of that. I generate well, maybe two generations ago in the 70s The death rate among child for childhood leukemia was pretty much 90 percent That virtually all children that got leukemia died And today it's pretty much the opposite. We cure 80 percent of kids that get leukemia And that's not better treatment. That's because we have drugs and tons of well decades of research effort have gotten into it I think that's the last real slide I had so for once I'm going to finish the time The study questions are intentionally a bit broader I will talk a little bit about this tomorrow that but I would like to let's try to split this about in A few challenges. So first, what are the fundamental scientific challenges to drug design? And that has to do with this thing that you need to find agonists, inverse agonists and agonists, etc You really need to influence something biologically Then there are a bunch of technical challenges. How do you achieve this? How do you achieve it cheaply enough, right? That I'm not interested what you can do for a billion dollars because you're working for My company that has a turnover of 10 million dollars. So you need to be realistic Then there are a bunch of practical treatment challenges that you can't have side effects You need small doses of protein people don't want to take an injection And related to that you have all these financial challenges. If you want 200 150 scientists abroad You need to be able to pay them. You need to pay the students if you're doing clinical tests And then I touched a little bit about how the fact we can use free end is try to think a bit forward looking here That's many of the methods we've gone through in this class because that's very much the future of the drug design. Yes Or the same company So this has to do with patents legislation in general It's complicated that a patent is ultimately not the right to make money A patent is a way for me to forbid you from doing something The reason why the society gives me a patent that might sound stupid But in return for getting a patent I have to make my invention public So that after those 20 years everyone would be able to copy it Otherwise you could imagine me starting a private clinic where you would come and I would give you the medicine in my clinic But you could never take the medicine out of my clinic to make sure that it stays secret and that would be a business secret is that So that of course if I am now and this has happened Losig is a perfect example so that roughly 10 years after the first generation They came up with the second generation of Losig that was more efficient. They pretty much just cut the molecule in half I bet they found that out earlier than after 10 years So yet you improve it a bit and then sure the original version of the molecule That is now public and can be sold by anybody but the new more efficient one That's more effective and that would you don't need as large a dose and we have fewer side effects that I still have a patent on And if you look at that, I'm not sure whether that happened to li peter But that's frequently that the revenue does not go down to zero when you lose the patent. So of course When the first generation of the drug can be sold particularly developing countries everyone where the cost is really Matter they will start to use the copy but the richer world They might actually prefer to pay you a bit of money for the slightly better version you have now The other thing is this repurposing Again, if I can take your drug and come up with a new way of using it. I can patent the new part But maybe there can be you might also be able to come up with the way of using a third molecule to do the same thing So that the problem is that it's going to it's harder and harder to make as much money eventually Yes, and it's also it can't be The requirement for getting a patent is that it can't be obvious So it has to you have to combine at least two things in a way that is not obvious to somebody skilled in the field So if there was just a matter of adding one hydrogen bond and do that was completely obvious anybody who looked at this would have guessed No, you won't then you won't get a patent in theory in practice. You occasionally get it anyway It's 3 p.m. We will meet tomorrow to tomorrow I'm going to go through The real science fiction future when we try to design proteins because these are small drugs Tomorrow I'm going to talk about making protein drugs And then you really need to understand folding and then I'm also going to go through some of these things where it can't go horribly wrong Because it has gone horribly wrong a couple of times recently And then the second half tomorrow I think I can have a q and a session for anything you might want to ask ahead of the exam or so But since you have two weeks or so until the exam I'll also make sure I'll schedule at least something online where you have a chance to ask question in the last minute