 Good afternoon everybody first fun to see you again. I was a little bit jealous of Burke and Lucy being able to give all the fun stuff about bioinformatics and everything But I've been traveling too much and I'm gonna travel a bit next week too, but there are two important lectures remaining today I'm gonna talk about Slightly more applied things, but we're gonna use all the things that you learned previous in the course in particular modern Drug discovery this is largely gonna be what the third-hand task is based on that I think that you're already able to download some of you might even have gotten started with it And then next week I'm gonna finish things up I will spend probably one hour during next lecture talking about modern protein design and where things are heading in general and Then the second half of that lecture. I will likely just do a recap of the course and the Q&A session one important lesson with that and that goes for people seeing the recordings to It's entirely voluntary, but I want I might spend 10 minutes or something going through the main topics or anything And after that it's up to you to her questions for me. I will stay as long as you have questions for me I will answer any questions But if you don't have any questions, I will happily leave the room five minutes after I get there So think about that for next week so that in case there is anything you want a question or whether we should revisit But the topic for today is drug design And in particular how we're gonna use this all the things you've learned about molecular modeling simulations free energies And I'm gonna cover docking in particular And what you probably picked up from the last few lectures is that this whole concept of Using the fundamental laws of physics the interactions that we covered in the very first lectures But then we rapidly got to the point where we realized just knowing the energies is not enough to actually explain what happens in chemistry And what happens in real life. We need to understand free energies And when you need to understand free energies, that's why we had to take this tour detour of first understanding standing entropy If you do not understand entropy, you're just fumbling in the dark when it comes understanding free energy And then if you're if you're happy in quantum mechanics or something if you only want to look at the energy of individual structures If you don't care about the world where molecules have multiple structures You can happily ignore and entropy the only problem is that that world is not the real world So we certainly approximate with a ton of things That's the whole reason why we threw out quantum mechanics early on in the course is that entropy in the real world Is usually a much bigger problem than the fact that we have slight approximations and our interaction functions and That's also why we got all the way to simulations the reason why we use Simulations that have just looking at structures that simulations allow you to capture the entropy It's not that you're simulating how a protein moves in real world and God knows that there I've seen my fair share of PhD thesis as opponent when you see that the PhD student happily claims that they can simulate How a molecule moves you can't and the main reason that you don't know the velocities to start with Second, even if you did know the velocities the Heisenberg answer to principle combined with the fact that these are chaotic systems would mean that they would diverge in no time But simulations are an amazingly efficient way to sample things and get a sense for the entropy and that makes it possible to calculate But what you would typically do if you work in a pharmaceutical companies roughly the sequence You would somehow get the new fancy smell see a receptor that they are super interested in because there could be a billion dollar drug And then they would argue that you should try to model this based on a bacterial protein I'm not sure how much Lucy covered that but the reason why this is so common is that it's very difficult to crystallize human proteins To tell the truth. We don't really know why human proteins are not very stable And bacterial proteins are much easier to work with and crystallize So that's why it's so common that we have a structure from a prokaryotic Bacterial channel like all these iron channels that she covered But then I would like to create a model of the human one so that I can try to design a drug because the drug design market for bacteria It's pretty limited. They have very bad health insurance But then they need to might need to build some side change you might need to Minimize the structure and then you might simulate the models of the structures But this will just give you a small movie about how the protein might move in everything That's not really anything that the pharmaceutical company is willing to pay anything for the reason why you're Interested in this it can we move things that you traditionally do in the lab and do them in the computer instead And that's what this lecture is gonna be about So in general pharmaceuticals is a complicated field that over a hundred years has been reduced to very very simple things There are lots of processes in the body Some of them are natural processes that happen and you get ill just because you get old and other cases There might be defects in your genome and other cases that might be Well, whatever external damage Say trauma or something and for whatever reason you would like to influence the process in the cell and as you might hopefully Recall from the very first lecture that if there is something happening in the cell and I said that there is a molecule involved You've heard guess should be that it's a protein involved So normally the way to influence a protein in the cell that is that we take this target And then we want some sort of drug complex some small molecule to bind to this protein and change something how it acts and this With the analyst at some sort of biological response if you're lucky The way that this is done for a hundred years is pretty much trial and error which is not as horrible as it might sound We'll get back to that The last 40 years that we've been a because we know no structures We've been able to unravel this in much more detail and we know for a whole lot of these drugs and pockets How they bind so you might for instance have a deep binding pocket Hydrophobic pocket where a drug is binding. I actually don't remember what that is. I should know but I don't This is another example where something is binding pretty much on the surface There probably there are a couple of questions you ask yourself How do you know that something is a binding pocket and how do we determine whether things bind there or not for now? We don't and that's kind of the problem, right? Both in the lab and in the computer. What can we influence it? How can we influence it? And in some cases there might be multiple different ways Normal if you have a structure that's nirvana But in many cases you're just working say on the biological level that there is something We have no idea what it looks like so just or it has a circle and then there might be something interaction with something else So in textbooks you will frequently see very schematic pictures like these and the only reason we know that these two molecules Somehow interact with each other and there might be some sort of Specific antigen peptide or antibody here, but we don't know exactly what they look like and then you just draw it schematically and Depending on how we change the look of these binding surfaces and everything we might be able to get either prevent things from happening or Turbo kick-started to make sure that it happens even without something present in the body if it's cancer for instance We might be interested in kick-starting the immune defense to make sure that the immune defense kills the cells more efficiently than they would otherwise do There are a ton of things in your body that we could theoretically hit There are the entire signaling system in and out of the cell and it's not a coincidence I why I asked Lucy to cover membrane proteins for you membrane proteins are primary drug targets There is a whole lot of things related say high blood pressure. That's usually proton pumps again membrane proteins a whole lot of neuropharmacological diseases Sorry neurological diseases neuropharmacology is the research about their The drugs targeting them And again, there's no limit, but if you start to classify all these Roughly one quarter of all the drugs target something called G protein coupled receptors And I'll cover with those are you don't I don't think I know and then I think this part is roughly Let's see. That's nuclear receptors and that's ligand gated iron channels And there's multi-gated iron channels now. We're above 50% of all drug targets All proteins targeted by any drugs and they're all membrane proteins With you could probably continue I would get that is probably two-thirds of them or so are membrane proteins And that's if you count in the number of targets if you count in the amount of dollars Which is pretty much the most important thing for pharma Then it's probably closer to 90% So while membrane proteins we frequently go around and claim that membrane proteins are important because I do research on them And they are important they account for 30% of all proteins in your genome or at least membrane associated proteins But in the pharma world, it's either a membrane protein or nothing Everything and that has to do with that. There are the windows and doors of your cells We are particularly into gas then they can get that iron channels. I might come back to that if I have time There are a couple of things drugs can do Many of the things I will tell you are actually based on history and remember historically until the 1960s We didn't have structures So how do you test things? Well, you would test it biologically in the lab Do you get the response that you were hoping for or can you prevent a response that is bad? And that means that you can try to classify things If this is the normal biological response and then you increase the concentration of the drug either You would expect it to Completely activate the receptor or whatever it is and I know this sounds abstract. I haven't really told you what the receptor is I'll come back to that But if a drug has 100% efficiency you get the full effect That's what we call all these drugs that stimulate that create an effect are called agonists And if you get the full activation, we usually call it a full agonist or sometimes just agonist And an obvious agonist would be the drug that normally does something for instance in the in the nervous system The signal transmitters their agonist they bind to the receptor and create a response You might have a drug That should somehow help this I want to I want to open the channels a little bit But you might not want to put the pedal to the metal, right? And then you might want to have a drug that's a partial agonist It kind of lubricates it a bit it creates a little bit of response But not everything and that would be called a partial agonist You would frequently in the literature sees something called an inverse agonist Hello, and then inverse agonist is pretty much a drug that creates the opposite response So this drug would normally open a channel this should close the channel is that Literally create something that's actually opposites And then it might sound very say why on earth do you have something that's called a neutral and antagonist that doesn't do anything well It's not that it doesn't do anything an antagonist that binds but prevents things from happening So let's say that you're a drug that would normally go and create That would normally go and open the door, right? My drug might simply block the door so you can't open it I'm neither opening nor closing the door myself, but I'm preventing you from doing what you would normally do in the body And that's a very Common type of drug. I wouldn't even say it's the most common one It's it's easier to destroy things in the body much easier to destroy than create an effect These are things you should be aware of agonists fuller partial antagonists and inverse agonists and in particularly the agonists and the antagonists are the common ones and Somewhere here you start to think it's easy You just you're gonna use simulations or docking as you're gonna do in the handling task and determine whether things bind and what their effect is And I wish it was easy, but I spent two three decades Doing research of that The problem is that part of the fact that that's difficult the really complicated part in your body is not what happens when the drug reaches a protein But it's everything else The reason why it's difficult is that the sec first you need to get the compound needs to bind to the target That's what I'm most you're gonna talk about today, but even if you have a compound, that's great It's not okay if it binds to 50 other targets because that creates side effects. There are some really horrible side effects to this We have one example of the ligand-gate that I in channels we started I think was murk I shouldn't blame them, but one of these large companies a few years ago They had a new compound on the market It went through all the stages of clinical trial and then at the very end when they started to test this on patients It turned out that a small fraction of the population had an allergic response to it And if you get that on the operating table, you basically die They had to pull everything from the market when they've already spent probably 10 billion dollars or something catastrophe And that's likely because it was binding or doing something elsewhere You must survive from administration that is eating it to where the target is the brain That is way harder than you think your entire stomach is based on destroying things right anything That's a protein is gonna be digested by the stomach You have a special barrier between the blood and the brain and the whole point is a few bodies to protect things from going Over this so that's super difficult You need to By basically if you want to what you call a blockbuster drug something can't sell it You want something that you can eat because no patient wants to go to the doctor and get five injections a week Well, they will if it saves their life, right? But you're not gonna go and get five injections just feel slightly healthier nobody wants to do that It's expensive and it's complicated that you're only gonna sell it to rich people even if it's really life-threatening disease And then you should have a slow and steady release of the drug So you don't need to take pills all the time and you want to get a little man What all this is called it was called ad metox absorption distribution metabolism excretion and toxicity? And in practice, this is usually the largest it's a relatively easy to find things that bind but tons of drugs fail here So what we've done historically is that actually that there aren't there is an infinite amount of molecules But in practice to fulfill these literally based on trial and error Lipinski several decades ago formulated that things have to be small So they can be transported in the blood and everything so they should have a molecular weight of roughly which is called half a kilo Dalton So roughly atomic units It should be relatively polar and you don't I'm not gonna ask you about what these log P numbers mean But it's polar enough to get into the bloodstream if it's a very hydrophobic molecule It might bind great to other things But if it's very hydrophobic, it's gonna stay in the stomach It will never be dissolved in the blood so it can't get out your muscles or brain or whatever you want a few hydrogen bond donors and a few hydrogen bond acceptors because if you have too many they're gonna start to stick together and everything and It should be a recent it can't be two polars It's a relatively nonpolar so it can get across membranes, which is kind of the opposite of that one, right? And that was true for drugs historically most of them would fulfill this the only problem is that the last 20 years Roughly we haven't seen a single new drug So that drug design modern drug design is kind of I wouldn't say I Wouldn't say that it's in a crisis, but it's getting complicated and one of the reason most drugs on the market They would never be approved today aspirin. Are you kidding me? That's a super dangerous drug you can overdose it And if you overdose it that you can destroy your liver and everything that would never be approved on the market today But you can go and buy it without the prescription Same with well a whole of these my normal painkiller drugs be careful with them They are really dangerous in particular in combination with alcohol, but because they're already on the market We accept that but of course requirements go up and up and up and We always like it when the authorities improve they have higher put higher standards for the drugs, right? But the only problem that's gonna mean that we have fewer and fewer drugs These are for examples of fairly typical modern drugs You have never heard these names Olanzapine, Pillow, Carpene, Psyllomethazoline, Chlamythidine and Cypressidone These are really the chemical names and then there are market names that are different all over the world So what the way that people have typically discovered this is that they have a particular protein They're interested and say that you're working with proteins related to high blood pressure and you as a company tend to Specialize on a handful of receptors and then you screen through and eventually you find something that you can refine and improve and turn into a drug So all of these are small quite hydrophobic They tend to have some aromatic ring structures. There is a reason for that. It has to do with entropy I'm not sure whether I'm gonna cover that later So I will actually say it now if there were no ring structures if these were large and very flexible molecules Can you imagine what would happen? Most drugs tend to be fairly small in rhythm So what would happen to the entropy when they bind if they are very flexible? Because then it would be much better not to be found right and that of course You could imagine having a super fancy binding scheme with tons of hydrogen bonds fitting and everything But this is all based on likelihood the likelihood that you will be able to find such a drug is very small So that's almost all drugs that you tend to discover tend to be have these aromatic rings So there's not to there might be one or two bonds that they look at the top one There's pretty much as one free bond that can rotate around so they should be fairly rigid So they don't lose too much entropy when they bind otherwise the free energy will never favor binding So where do you get these from? divine inspiration Pretty not actually not that far from it. So first it's important divine inspiration would be okay And I think I covered that a little bit before in the course that there is nothing Actually might not have there is nothing that says that this for instance is the best anti-psychotic drug I can virtually guarantee that it's not but you don't need the best one You only need a good one So it's not like in nature where you've had four point three billion years of trial and error and need to find the refine The absolutely perfect structure. We are very happy here as long as we have something that has some sort of effect so the traditional way of getting these is pretty much from rainforests rainforests is just an example, but you tend to find something in nature and say coca realize that there is some sort of Tribes or something and they have a habit of eating leaves from a plant and if they eat leaves, they don't get tired and Then of course you would spend a few years researching this and then you would identify the leaves and then you would try to identify What is this chemical in this particular plant that has this effect? Or it might be a beetle that if it eats something it's resistant to another poison or something and then you try to identify What is that causes this effect? Or same plant or whatever And then this might not be very efficient and then you try to imagine could we create a drug? That looks roughly similar to this isolated compound Can I then design a compound to maybe I might not even know what the receptor is But if I know what the receptor is, maybe I can use the computer to design something that would fit it even better and Sometime in the future we would like to be able to Design arbitrary peptides proteins and the this is still a bit of a pipe dream The reason for this these small molecules don't have a whole lot of freedom They're fairly simple with proteins. We have all the tools in the toolbox. We could design anything We could literally tailor make a key that would fit only your receptor, but that would not hit anything else in the body The problem is you have something small and hydrophobic, right? It will bind to my receptor the likelihood that it's not going to bind anywhere else in the body is fairly low And that's when we get side effects. So in theory proteins hold great promise, but it's we're not quite there yet The way this works both historically and now is that first you need To identify a target and that's that might sound obvious But remember the last slide on the rainforest I know that if you eat these leaves something happens, right? But I still have no idea what response does that solicit in your body? So in many cases you want to identify This leaves or whatever it is what protein is that binding to is that might be a membrane protein or something Or angiotensin receptor if it's related to the blood pressure or something and if you identify these receptor We might start of a few candidates about what the binding sites might be And at this point if you're a run a large pharmaceutical company You might even determine the new structure of the protein to find out what the binding site is and Then we need something to start with For now, let's say that that's divine inspiration. I'm going to come back to that shortly and then we need to see does this have any sort of effect whatsoever tested both in the lab and In computers and everything in general, that's going to be a very bad effect Why it's bad I'll tell you in a second So then we need to improve it and at this point we usually go what we call hits to lead Here we have a clue things are just pulling the lead and At some point if we're happy with this you're going to start doing tests first in the lab and then eventually on animals And somewhere here the researchers like me We are so thrilled that we completely give up the idea to a pharma company things have hardly started yet This is where it starts to get expensive So then you need to do three sets of studies the first one is you need to ask is it safe in humans? the second thing you need to ask is it Efficient in humans That's a completely different question. It states one or we want to know is that it doesn't kill you Does it have any effect whatsoever on the disease we're trying to treat and in phase three? Well, say that you want to treat blood pressure. There is a whole range of drugs already on the market We're not going to prove you new drug on the market unless it's better than anything. That's already on the market And what you don't want to do you don't want to fail here because that's when companies go bankrupt because it's astronomically expensive The problem with this is that you fail you fail all the time Actually, we researchers so we fail something like 70% already before you go into the clinic. Is that good or bad? Perfect answer, but it's a good or bad Yes, it's awesome. Why is it awesome to fail there? Because it's far better to fail there than failing here. This is the food and drug administration Here is where you have invested ten billion dollars of your stack all this money. This is where CEOs get fired So that but this is where it cost of one million dollars Yes, and researchers had a project that we decided not to pursue it because it was too uncertain. This is awesome So you want to push failures down here and this is where computers come in The point is not for the computers to predict the perfect drug the perfect drug we will find out here But the computers might help us to turn that you know what in all likelihood We should not pursue this because it's uncertain whether we will be able to do it and computers have the advance We can have very very high throughput So that the typical cost of developing a drug might be take between 10 and 20 years It takes well time to patent is a bit uncertain It might take a ballpark of 15 20 years before you can patent and this is a problem because your patent is valid Roughly for how long? 20 years So if you get it to market up to 50 years 15 years you have five years of protection remaining There are some exceptions to this where you can extend these patents and everything but This is why drugs are expensive You have invested all that money and you have five years to make the profit back after that is free for everybody to copy The cost might be in the ballpark of half a billion dollars or euros For a normal drug the most the most advanced ones are more expensive And it might require a team of 150 scientists or so sorry General roots for patents That's an interesting question person. Why do we have patents? You're engineers. You said this is important, right? This is goes to the core core of what you're doing Why do we have patents? No? That's what everybody thinks Sadly a few whole other politicians seem to think so nowadays the original reasons for patents is to provide an incentive to share an old Because the alternative patents is to keep your invention secret, right? But in to the point of when you have a patent and it's issued it's public But in return for making it public during the next 20 years You have you are the only person allowed to manufacture this product and I have a right to sue you if you try to copy me But again the return is it off to 20 because you made it public right away, right after 20 years Everybody can do it So there's this balance between the invaders individuals right of making a fair profit But also the public's right on being able to build on knowledge So that if we didn't have this everybody would just keep their chemical secret and you would probably be able to go to a Some sort of clinic in a hidden location. You would get a drug and nobody would tell you what the drug is So there are a whole range of tools you use in these studies So if you have in the very early first to identify these targets and everything There's a whole lot of bioinformatics actually we try to identify sequence variations We might want to understand why is that you are susceptible to disease that most of the population is and can we find deviation in your genome? Ten years ago. That was science fiction today. It frequently works At some point this middle part high throughput screening here We do a lot of computers are quite a bit on the lab, too But this is becoming more and more computer driven and eventually when you go into preclinical development This is doctors and clinic and just injecting things and testing lots of statistics So if we go back to this plot that for me This is intermediate part is really the interesting one finding hits seeing whether they have an effect and trying to optimize things And this is where computational tools are Ten years ago, I would say that they start to make a difference today. I would say that they're driving everything So what we typically happen is that we would have some sort of initial discovery molecule and I would identify this Based on some sort of database and then hopefully I have a series that there are four or five of these that look roughly the same that these Are interesting ones and at this point literally mildly interesting is just a keyword This is not a typical drug company will have hits every week And it just means that it just barely rises above the noise level And you tend to have an iteration frequency of four weeks every four weeks You want new results both from the lab and the computational team and everything and then we see have we improved The interesting thing is of course, can we improve and make this better? And the way you do this is Typically that we test things we need to test things at insanely great scale And that we typically do in the lab. So these really cool machines. They're super expensive They can pass something like hundred to two hundred fifty thousand compounds per day And they just have these gigantic Microarrays and then you're testing different drugs under different conditions And then we're screening this through with the computer and seeing in what cells did we see binding did anything happen? And it you pretty much just want to plus zero or minus sign You don't care about the effect or anything because again with hundred fifty thousand If lucky you might get hundred leads or something The likelihood that you get zero leads that that's not gonna happen You will always find something which is the problem because that these leads are not worth as much as one the cost is fairly low Maybe a dollar per well or so Until you start looking at Hundred fifty thousand of them per day This can be fairly expensive And hundred fifty thousand is nothing imagine the size of chemistry space right all the possible drugs you can do There is no way you can test all possible drugs And the first testing this even assumes that you have the drugs in the first place Synthesizing a drug can cost fifty thousand dollars if you need to synthesize a new molecule and higher organic chemist to custom design a molecule That's not gonna happen So that in general that there are tons of ways that we can and there are other experiments You can use polarized lights and everything but anything that can test whether something happens can be used on some sort of screening In principle this shouldn't really work because a chemistry space we sometimes talk about 10 to the power of 60 or something molecules Which is just estimate The probability of finding a ligand that binds by random screening Just one million of those that you're testing one molecule out of 10 to the power of 10 The likelihood of finding a good drug is literally zero but it's Sometimes it works There are these are examples from a few studies that a friend Friend of mine in Uppsala. Yes Carlson was involved a few years ago the two targets It doesn't matter what they were they tested three and two hundred thousand compounds respectively and the first one They got zero experimental hits and that's when you cry And the second one they got almost 150 The problem is that this expensive imagine now that you would like zero is not a good number to return to your boss Would you expand this to three million? 30 million At some point they're gonna say okay, we do 30 million, but if not I want your resignation next if it doesn't work I want your resignation We're saying in 30 million times $1, right? That's That starts to show up in the books So what you would like to do is can you do this more efficiently in a computer? And that is what docking is about I'll get back to that in a second So what you would like to do in docking is using sort of large computer And Instead of doing this centers for you the largest possible centers in the world might be able to do hundred thousand or maybe a million compounds But if you do this in a computer, you might be able to bring it to a million per day or maybe a billion for a total project And if you're testing one million compounds while my company is testing one billion compounds Sure, my method might not be as accurate as yours But both our methods are very approximate to start with if I test one billion compounds I'm gonna have a 1000-fold head start on you And I have a room for lots of mistakes along the way So the problem here is you need to be fast sloppy, but fast fast fast is the keyword and I don't care so much about accuracy These are just random Literally think of this is divine inspiration if I come up with something. That's good. We'll test it There are a couple of very simple ways and in particular this is machine learning is but this is actually machine learning method One of the simplest ways something called QSAR quantitative structure activity relationship, and that sounds very fancy, but it's not really Did Lucy talk to you about the ligand gated channels and anesthetics This is my love in life because that's my research Anesthesia is 150 years on trial and error. It's not really until the last 20 20 years So we've started to understand what happens when you fall asleep But for over 100 years ago my earned overtone independently formulated an hypothesis that said the more Hydrophobic a compound is the better. It's going to work as an anesthetic That sounds like a fairly simple test, right? So then you plot the oil to gas partition coefficient and Against the potency of an aesthetic. I don't know what's going to talk about how you measure that right It's a pretty darn good correlation from zero point one to one thousand So this covers full orders of magnitude. That's the type of plot. That's good when a student send that in in the lab You're convinced they cheated It's perfect So they drew the conflict and of course if I now give you a new molecule that has this No, sir This is more hydrophobic and this is better So if I say if I have a company that has an oil to gas partition coefficient here, is it going to be good? Good is down as an aesthetic or is it going to be bad? These are the good ones. So if you go even further down here, do you expect that it's going to be even better, right? Yeah, not exactly rocket science So that's basically we're saying the the structure here or the the structure or the quantity is based If it is hydrophobic, we expect that it will be a good anesthetic. That's pretty much a linear correlation. Yes But I don't think that that's a poet. That's not a potency. That's probably the maximum Maxim alveolar concentration so that the concentration you need for it to work Well, and the reason why I can say that that carbon the nitrogen they're kind of okay But I recognize all these isofluorane chloroform. These are the good anesthetics. That's why I said that forget about the y-axis there And you could do this slightly more advanced You could calculate the number of expected hydrogen bonds if we have an aromatic ring if we have a hydrogen bond donor Close to an aromatic ring. Is that usually good for this particular receptor? And what you would do today is that we would throw all this in a machine learning algorithm and AI is super popular in this area now And literally you forget all about physics and it kind of works sometimes and sometimes not The advantage is that it's super fast because once you've trained this you can predict probably not just 1000 You can probably predict a million compounds per second So you can take not 10 to the power of 60, but probably 10 to the power of 12 and just screen through them Those predictions are of course going to be lousy But I will come back to that in a second that The point is not that they're good The point is that your predictions might be slightly better than chance and that's frequently enough You could tell you the mass the molar weight the charge the dipole moment and everything the Advantages and it's fast the disadvantage is that if you make a very specific model That I know that I know that for this particular type of receptors I need an aromatic ring and I need to hydrogen bond donors. You kind of already set the rules of the game, right? So you will find lots of things that have one aromatic ring and two hydrogen bond donors But you won't find other things that might be even better So that's literally was it's an okay ish method, but it's no nothing that makes people Right home and say that that it's used everywhere, but it's never really the deciding factor But still if if I happen to have a compound and a target that I know really well You could sign of rather than having the drug maybe I can make a blueprint of the drug What is rather worrying about all those atoms just as you used an alpha helix or beta sheet to do describe the Important properties of secondary structure. We could do the same thing with the drug So maybe it's important to have an aromatic ring there an aromatic ring there and then two aromatic rings there And you need a couple of specific charges or hydrogen bonds But instead of having all the atoms you can describe the same There should be an aromatic ring there and then maybe something hydrophobic there and hydrophobic there and then say a dipole There and then just measure all the distances between these and then throw this at the database and say find me other things That looked roughly like this and this type of abstract model of roughly what things am I looking for in my component? And what distance is from each other is called the pharmacophore? And if you're going to work in the pharmaceutical industry, you're going to hear this because they're I wouldn't say that we're they are obsessed with Pharmacophores, but during the drug design this kind of the typical ball print I'm looking for something like this, but it should be even better and Then there are large databases where things like these are described and again It's pretty much just machine learning find things where the set of distances is roughly the same as what I had and then there might be When you do that you typically turn out that there are a bunch of common elements I won't go through what this drug is but so this is a particular drug that you had a full agonists and Then you had a bunch of variations of these and they're all different agonists to a greater or smaller extent And here you can probably start to see all of them tend to have these two aromatic rings in the middle Right. Well, that one has an aroma has slightly different rings and that has slightly different rings But they're kind of similar, but you have different so-called Substitute groups at the site And what you typically do in the lab is that you sit down and do more or less by trial and error We said well, you know if all the good ones had a sulfur up here, maybe I should try to add two solfers Let's see if that works better or if all the good ones had hydrogen bond donors here Let's add a second hydrogen bond donor or make it more hydrophilic in this part or more hydrophobic in this part And the second you've done that what do you need to do? Guessing is easy. I can easily say that let's add another sulfur then you need to pay 50,000 dollars To have this drug synthesized because now you need an organic chemist to make your specific molecule with one more sulfur And you need to produce maybe a milligram of it so that we can test it That's fine for the first 10 weeks the 11th time you come with a new guess and it doesn't work The the person in charge of this that it might start to get a bit irate, right? This is getting very expensive and That's why we would like to instead of synthesizing these so that you can do a real test What if you could just throw it in a computer and let the computer say is it likely to be good or not and that's where we get There are a bunch of things like volumes that I'm not gonna be Go into here, but what I want to get to before the break. We have a couple of minutes is I'm gonna stop myself I Have completely ignored the protein structures that the rest of the course was about And we're not gonna in some case you might not need drugs Very structures of protein to do drug discovery if there are lots of existing drugs already I might just try to learn from pattern recognition if you are for pharmaceutical companies I might look at the drugs that you have produced and then I might do what you call a me to drug And a me to drug is that I try to copy the properties of your drugs and make something that works around your patents So it should not intrude on your patents But be similar enough that will have the same effect and that's awesome because it just takes me a year to develop that drug And then I will steal one fifth of the profits in the market But I wouldn't be saying this unless we needed the protein structure and that's where most of the things are happening today molecular docking so You have seen a bunch of proteins a particular member of protein in the course and if we have the protein structure We might in some cases with or even structures with known drugs crystallized them So we know that in this particular and this is dopamine receptor. We know where the dopamine molecule is found We might even understand the pharmacophore exactly But if we have the structure, can't we just take other molecules put them in a computer and see how they might? The good news, that's it eminently possible You could even do it with a molecular simulation. The problem with that you've all seen how slow a simulation was, right? You can take one molecule and it would take you two weeks So what we want to do we want to do things like hundred million times faster And that goes back to when you were like six months old. We're gonna do roughly this So you're just gonna test things test things throw it away If it doesn't fit you throw it away test throw it away test throw it away test throw it away and do that ten thousand times per second You're gonna be throwing away a whole lot of stuff that is really good But I don't care because there is more than one drug in the universe All I care about is reducing the number of bad things I have to test in the lab It's slightly more complicated in reality But not a whole lot. So they they're really it's a very simple question You want to find out what is the best ways to put two molecules together? So first define best Well, we're not I can't I'm I'm a poor bastard here So I'm not I'm not going to be able to use all those fancy interactions that we occurred earlier in the course water 12,000 atoms I can't afford water So we're gonna need to have some super simple ranking solutions. Just base of two polar parts are close together. That's good I give it a plus one if two If a polar and a hydrophobic things are close to each other, that's probably bad. Let's give it a minus one So you just find some ways of putting a score. It's not really a physical energy If I have a very flexible molecule and that has to become very rigid. It's losing a lot of entropy. Let's say that's minus 10 Completely arbitrary and we can we can try to calibrate these based on experimental results or something But come down with something that you can test in a tenth of a millisecond You could use a force field, but that's usually too expensive But then I also need to test many ways of putting these molecules together I want to test the molecule in different orientations There might be more than one binding pocket Maybe the protein can move a bit. So for every molecule I might also want to test it in 1000 different paces of each receptor And now I already used 0.1 seconds for one molecule So throughput throughput throughput everything here is about throughput So we need some sort of good search method And what you typically do is that you come up with a way of sample things Generate lots of conformations of the small molecule and this can literally almost be random I would even say that it is pretty much random for several algorithms And everybody's trying to argue that their randomness is slightly better than Charles And then we just score them and then try to keep the best ones If you happen to find a score, that's good Maybe we should spend a little bit more time on this molecule more do more fine-grained sampling But again the point here is not to do it perfect The point is to do this fast and sloppy And use the best way to screen as large as a part of the chemical universe as I can We have two more minutes So even if you do this in the most horrible ways we can imagine There are like six rotation and translation degrees of freedom, right? There might be four bonds inside the molecule that can rotate That's already 10 degrees of freedom And then the box let's say that it's just 10 by 10 by 10 angstroms I want a sample And then I want to sample just this is like live and pass paradox Just come up with some sort of steps Maybe every half angstrom and say sample angles in 10 degree intervals And if I could do 100 conformations per second It would take 200 years to finish that for one molecule This is unfair because in principle it would probably be more like a hundred thousand per second So I could probably finish that in a month or so But this has everything in docking is focused on speed And what you then do is some You've probably seen this in optimization But we need to make some sort of initial population randomly either of many molecules or the states And then evaluate scores and check which ones looks promising Either in the sense what molecules look promising or what poses of molecule The places where it by and look promising And then we try to extract the best ones But my survival of the fittest right And then spend more time there maybe test that in a more fine grain motion or try 10 different versions of this class of molecules seems to be interesting Let's pick another 100 million of these class of molecules because we might something interesting there And then we do repeat it and repeat it and repeat it If you can come up with a better optimization algorithm That will work equally well I will have one more slide and then we'll have a break You could also cheat because we don't have to obey the law of physics I could take my small molecule here and then try to grow the molecule here I put my first aromatic ring and then I see can I fit a second aromatic ring here? Nope, I bumped into the protein. Okay, throw away that molecule Maybe here grow the first aromatic ring here the second aromatic ring here that worked third aromatic No, that bumped into throw it away So that any way that you can come up with something is awesome So this is surprising this is way more of divine inspiration than you think if you can give me 100 molecules that are worth testing in the lab Those molecules we might want to spend $50,000 each one And then we can come back in four weeks and see what is the experimental result of those molecules and based on those tests We can then decide in four weeks. How will we do next batch of computational studies? But I will come back to that after the break and talk more about scoring and a little bit more practical examples So let me the quarter past should we resume So I spoke about these ways of trying to well more or less by divine inspiration come up with algorithms that find things that are at least Compatible actually I would think compatible is a great keyword If things appear to be compatible, let's put it on the shortlist and retain it But if things bump into each other if for whatever reason it's unlikely that this is going to be good We take it off the list And this comes back to what I said failing early is failing cheap So it's much better to In all likelihood, we will likely remove some very good hits And that's fine because it's very unlikely that that is the only interesting hit And as long as we find something of the say 100 molecules retain or something that's going to be fine And Similarly the way to get this fast is that you could use the force fields we covered before there is a slight move and I'll come back to that Of actually even using simulations But most of the things that we use very empirical scoring functions that All the things you've learned in the course when we know that it's good Hydrogen bonds charge interactions or hydrophobic interactions We just try to very arbitrarily give that good scores and in theory we could also use some statistics and And as you can probably imagine this field is exploding with artificial AI based methods and machine learning to nowadays And I'm not gonna I will skip this grid because it's not important But in principle just come up with any type of algorithm you decide how to sample things So here we're not trying to reproduce physics We just want to see what is possible and what is not possible and save the things that are might be possible So if we go back to that table I had for both of these compounds lack the mason crude same We actually got two and five Docking hits respectively. So these are completely computational hits And yeah, those five hits compared to those hundred forty six experimental might not sound so cool The advantage is that this might have cost you a hundred dollars Kind of nicer to spend hundred dollars and two hundred thousand dollars, right here Well, this might have also cost you a hundred dollars and you got two hits Which certainly beats spending a third of a million dollars and not getting any hits And at this point they might certainly be lousy, but I don't care so much that they're lousy This is something that to start working with and as long as you have a starting point. It's much easier There are as computers have gotten faster and faster. That's the other important thing is that these experimental methods We can certainly buy two of those machines, but then it goes on the cost twice as much And anything experimental in the lab and it does not develop that quickly But all the computational methods they've become roughly twice as fast every year Because computers get faster all the time So what we've been able to do the last few years is that you can include say receptor flexibility or at least flexibility of the molecules So you might want to test the protein both in the closed and an open state or allow the small molecule to push some side chains apart a little bit Because just maybe maybe that might make it fit better And at this point you might actually have a drug This is something that binds You will have to test it experimentally too, of course But if in many cases when you test it experimentally it actually turns out that there's it's not a perfect correlation But there's certainly a correlation that things that we tend to give low scores in the Docking definitely tends to bind a bit The only well there are some minor details Such as you would eat five kilos of it per day The problem is that binding in chemistry is an equilibrium, right? And you don't have a very good binding coefficient So the only way to get enough effect is to add so high concentration of this molecule That you would push the binding very much towards the bound state You can imagine what's going to happen that five kilos of medicine in your body that's going to lead to a ton of side effects Forget about it. It's useless And this is the problem so that to get rid of the side effects It's a first to be able to get rid of the side effects and second to make sure that you can actually eat one small pill a day You need to have insanely efficient binding So it should only bind to this thing we wanted to bind to and ideally nothing else So there is a bit of homework to do here. We need to improve this by a few orders of magnitude And one of the first Actually the first famous example was the HIV one protease inhibitor I know you're not chemist here, but I'm I'm gonna need to use a word So this was a first hit that they found in docking which is symmetric diol that had some tiny activity And from that they designed a pharmacophore. That's as roughly what are the three parts you needed of this molecule And then we found that hit in the database that looks quite different from number one And to make that simpler and having fewer flexible things at some point the chemist decided to remove a couple of things We had a simpler molecule that was the initial design And then based on a number of computational screens and everything you came up with an existing design with the diol so they had two alcohols And then we added some urea groups here And then we optimized the stereochemistry for binding. Sorry, that was urea Here's the optimized stereochemistry for binding and this was the final drug And that was one actually not this one of it was the first drug that was used in HIV treatment And it's famous because the first drug that was actually designed computationally There were certainly a bunch of Experiments along the way here, but you see the contrast amazon This was not based on the fact that we found a drug in nature that already had this activity We started from a computer. We want to do something that inhibits a specific protein the protease Can we find something in the database instead and let the computer do the early part instead of the rainforest? And it worked and since then there have been dozens of more drugs like this And this now enables you if you're a pharmaceutical company interested in blood pressure, right? You don't need this lead from nature. You can now go into the databases and computers is that which opens completely in your worlds You know what this is This is something that makes a fighter jet look very cheap. This is 140 billion dollars of revenue One of the largest and a lepetre which is an drug used to treat Cholesterol basically Is a great lifestyle drug This is the type of drugs pharmaceutical companies love because you tend to give it to rich westerners that have lots of money They need to take it the rest of their lives At its peak it had revenues of over 14 billion dollars per year and until this 140 billion dollars figure was on till 2015 And again, this is like this can be an order of magnitude more than like like companies say like saw the selling fighter jets for And this is why there is such an industry. There is an amazing worldwide industry for this This if you go to 2003 until 2016, this would happen. This is revenue. What happened here? This is why the drug cost so much right these were the golden years This is where they put it on the market and in this case they there are some patent extension laws now So they had slightly more than five years But this is the period where they could recover all their investment costs And of course this this drug was that it was an amazing Return on investments, right? But for each such drug, there's probably 50 or if not 100 that failed And of course you need to recover those costs too so that this certainly Make no mistake. They made a lot of money here, but they're a very short span. You can make money at this point It's all over that you need to restart so what What we if you're better more efficient you can find better drugs You can find drugs that are get on the market earlier Imagine if this drug had been on the market three years earlier You would have three more bars here corresponding to 10 billion dollar seats. That's 30 billion dollars That's worth a bit of computer time So both in docking and everything we would like to use simulations and everything to make things more accurate faster And in particular avoid avoid the needs for lengthy experimental trials, but also if we're going to fail It's much better to fail right away And Burke I think already spoke a little bit about free-end your calculations and simulation and everything and in principle We are really good at calculating binding molecular dynamics can give very accurate results The problem is that it's it's slow. It's expensive computationally not in dollars, but computationally there are A few things that have happened the last few years. One of them is this david shul. Did loose your Burke tell you about david shul so this is a guy Who was a professor of computer science? I'm actually going to show you first This is a brute force simulation of drug binding So you see that this protein this small drug here is eventually going to find the binding pocket in here And I think this is a simulation that covers roughly well several hundreds of microseconds It's a couple of thousands times longer than anything that you were I we did so david shul Was originally a professor of computer science But then he went very early on left academia and went to morgan stanley fair correctly and pretty much the father of modern arbitrage trading in stocks Because he's a computer expert and then some 15 years ago. He went back to academia more or less as a hobby But of course if you run I think it's at its peak the d the d show Company was the world's third largest hedge fund So that at this point you don't apply for an assistant professorship But you just hired well he invested a couple of billion dollars. I guess it's to hide 50 people for his toy company So they've started to design custom hardware ASIC application specific integrated circuits that can run simulations a thousand times longer than anybody else to really go Off to these problems And one of their goals is of course that they would like to be able to predict drugs and sell these to pharmaceutical companies and today this is still super expensive, but When I was your age, we couldn't even dream of this. So give this another 10 or 15 years. This might very well be the norm in pharma And what you can do then is that all the things that we've looked both in terms of protein structure and simulations We can start to understand any landscapes We can start to look at depending on where the ligand is situated What is the interaction energy in this case? We see if the ligand is close to the binding site the interaction energy is low And I guess this is another actually there's two examples. Sorry. You don't see it very well here But the overlap here is The gray structure is the x-ray structure While the brownish structure there is the structure coming out of the simulation Without any knowledge of the x-ray post So the simulations have been able to actually predict Pretty much exactly how the ligand should bind Without knowledge of experiments The other thing that this can do that In theory we can do this with docking too, but in some cases the problem is not just getting the molecule to bind But the question is how long does it take to bind? Remember the thing I said about kinetic versus thermodynamic stability, right? And just as for the protein folding it's equally important here The fact that in theory you should be super stable in here that doesn't really help if the molecule can't ever get there Because it would take six weeks for your drug to take effect. You would be dead if it's a lethal disease So in many cases there are kinetic barriers that it takes a very long time There may might be a native post in there in the weak Gray structure while what we end up with in a simulation is really something that sits out here It hasn't really had time to go all the way in because we need a bunch of waters and everything need to move out The waters there end up being trapped and that's if you wait a very long time Eventually those waters will diffuse out and the ligands might take its place And this can be important if you actually want to optimize what you call the onset and offset It's important that the molecule actually binds quickly This is super important in anesthetics So It's fun once upon a time I was hesitating whether I should go into physics or medicine and that's probably the hardest decision of my life And I feel that everything in medicine Sounds interesting everything except anesthesia because being anesthesiologist must be the world's most boring job And then I've spent the last 20 years working in anesthetics It's actually it's super cool Because it is so intimately coupled to What we mean by awareness and consciousness One of the difficult things and I think the reason why I need to especially dedicate a doctor and anesthesiologist is that it's difficult It's super difficult We basically we need to take you to the brink of death but be able to revive you So first you add an anesthetic that makes you fall asleep But of course this anesthetic that makes you fall asleep At some point it's also going to mean that you have muscle relaxance and everything and your lungs will stop breathing That's bad because you're going to die So then we might need another drug to either keep up your heart rhythm We might need to ventilate you that makes have a machine breathing for you The only problem is that that might do something else to your heart And then we need a third drug to compensate for that Then we also need a drug to make sure that you should not have any memory of this So that's a fifth drug you'd like ever giving this really complicated cocktail of drugs And all of them will have different fall-up times But they might 10 take 10 minutes to fall off. So imagine that I sedate you And then I've given you a little bit too much of this drug that oppresses your heart rhythm That's fine, right? I could just reduce the dose the only point. Yeah, that takes 20 minutes In 20 minutes from now when you're going to be dead So then you need to add a sixth drug to maybe keep up your heart And then it's fine. We're over with the surgery It went well, they stitch you up and they roll you out to recovery and now there are six drugs and they all decay at different rates So what if the drug trying to keep up your heart rhythm decays faster than the one depressing your heart rate? Then you're going to die in recovery And that happens people do die in recovery So it's also about we need to understand not just that it binds but how quickly they bind and unbind And that's very difficult to get for talking and anything and simulations are starting to play a role there I will come back to that to give you the examples because remember what I started saying deep protein couples receptors These this is a gigantic family of proteins. There are 900 genes or so in a human coding for them. Not all of them are expressed This is a bad way of saying when I was your rates the number of known structure was around number zero We did not know the structure of a single deep protein coupled receptor And everybody said that oh, it would be great in the future if we had one But the likelihood of ever seeing one is probably zero because the rumor at the time was that there were companies that had Been two billion dollars trying to get structures, but nobody had succeeded Today we have 14 so we know a huge amount of the tree of these The they are very difficult to crystallize that's the main reason So what happens is that you get some sort of neurotransmitter or something that binds in the extracellular part here Then a miracle happens The entire structure of these receptors changes through the membrane And that in turn leads to a single being transmitted to this g protein that sits on the inside That's what's called the g protein coupled receptor This is the receptor that binds the small molecule and it's coupled to the g protein And when this undergoes a transition it basically signal things on the inside of the cell So this is the telephone network of cells The reason why There are many reasons why it's popular apart from its importance It was also one of the this class of proteins was one of the early membrane proteins because it's very simple It's seven helices that go straight through the membrane So it was one Rodopsin and bacteria rodopsin in particular that's closely related We're among the first proteins that we learned structures of And the cool thing is that some a little bit over a decade ago There were suddenly two structures published in nature the same week of the human beta 2 adrenergic receptor By brian krabilka and ray stevens and I have to confess that I'm very partial here because I was a postdoc You're not in brian's lab, but in the neighbor lab Rumor has it that they were even some collaborations between them But that they ended up being a split and then they fought and basically rushed to try to publish this boat I'm not sure if you know the end story of this but in roughly five years later brian krabilka got the noble price for this Together with his advisor for a bit specifically for the studies of the deep protein coupled receptors not for destruction So ray stevens was left out there The amazing thing with gpcrs is that if you start to look at the within there's a tiny binding pocket out there And even to me that it's virtually impossible to see where things are mine It's actually not only impossible to see even if you have the x-ray structure if I don't see the ligand It's virtually impossible for me to say where the ligand is So it's a tiny binding pocket that is also very diverse different tpcrs will bind different things And again, this is amazing if you have a telephone network because this is literally different extensions But if you want to be able to understand one receptor based of another it's really painful there are One of the examples of this receptor. There is a Carazzole this particular is a partial inverse agonist That is a beta blocker that it's basically Protecting the heart from the second heart attack once you've had a first one And as I mentioned there's been an explosion of structures here So first you have the high resolution structures and you get the active state structures You know the structures with flat first receptor and g protein complexes. We now have mmr structures Cryeum structures and everything so we're starting to understand not just what the structure is but the entire movie What is happening as it undergoes different states? What's happening? What the molecule is binding? How is the receptor being activated? Which again, it creates in a completely new universe where you can design specific drugs in it And we even have structures with lipids Some of these structures were actually initially on only available to companies And I think that's what how some of the some of the first structure with Ray Stevens He funded the work that way that there were companies willing to pay for having a one-year head start on everybody else So the structure was deposited later, but during 12 months only a company had access to the structure And then you see more and more and more receptors David Shaw has been involved in this too so I'm going to show you a small movie of a Have a look at the time scale there and try to remember that in comparison to your simulations And this is the small neurotransmitter and then we're going to see what happens with the molecule here We're already at half a microsecond here And then it's going to start to bind in the pocket And then it goes deeper down and then it goes even deeper down Do you see that these were pretty much kinetic things that it had to get over? And now we started to expand the time scale so that after roughly five Microsecond here so that either start to push the helix out So now the entire receptor here went through a bit of a structural shift You might see it better if I go back. You see that was the initial state And what this is the entire structural shift that transmits a signal to the g protein that sits on the inside That's going to create signaling inside the cell Not only that they were able to get a ton of data about this binding site So all the random motion out here eventually you got the ligand down to this inner binding site So let's compare that binding site with what we actually had in an x-ray structure Purple one is a simulation Gray one is a x-ray structure Pretty sexy, right? So the computers are able to predict not just simple chemistry We talked about earlier in this course we looked at hydrogen bonds, right? Or predicting how oil and water separates This is pretty darn difficult. You didn't even see all the lipids and everything It's a gigantic system. It's a large molecule that undergoes a major structural shift It's a drug that we know is not only pharmaceutical irrelevant We talk about a lot of money here and the computer can at least within a couple of weeks predict exactly how the drug is going to bind And based on that we can even if based on these simulations You can even estimate what these barriers are just like the ones we drew schematically before in the course What is the first kinetic barrier and then you get to this vestibule part And then the second barrier when it jumped to the second step And this corresponds very beautifully to different states we see in the simulation And I'm let's see You can do similar statistics and let's take a binding site and compare to what are what are the things that we are close to What amino acids are we interacting with and if we now realize that there is let's say that there's one or two amino acids here or contacts that are This is not true for gpcrs But let's assume that there is a mutation here and that we know that people with this mutation tend to be susceptible to particular illness So maybe would in the future Maybe we would like to do a drug that's not just a general drug Can we create a drug that specifically combats that mutation? So that if you have this mutation, it's going to be a very specific drug for you You want the other and have a different mutation? So when you get this disease, you can have a slightly different profile of the disease But maybe we can have a second variety of the drug that is tailor made for people with that mutation So this you might have heard this concept is called personalized medicine and it's one of those People are spoken about this as the next big thing for 15 years. It's still the next big thing It hasn't really showed up the problem is I think it's how to test this How to make sure that it gets Well, it gets clinical approval and everything and you still have enough time to recover the investments before the patents expire But there's lots of potential improve revenue here And well, I talked about revenue. Of course, there's lots of potential impact for very severe diseases too I think that There is a second orthosteric site. Let's see if I had a full movie of that one too It's the second movie You see there's a second molecule binding in a slightly different site And here we're already up to seven eight nine 10 micro seconds Pretty long simulations. I think your longest one was hundred nanoseconds or something They've done simulations up to a millisecond not of this particular protein And the cool thing is on these time scales This was pure science fiction when I was a PhD student But on these time scales apparently it's long enough that we can actually start to see biology The roughly where the state of the art today is that the second part is that we would like to understand the entire activation here What happens when you're binding things in a large protein? This point here is not so much the deep protein coupled receptors But I will still cover this as a way for you to see the coupling between physics and biology The complicated thing in biology is that what I just showed you is a horrible lie and over simplification In biology, you would have the small agonists the ligand binding You would have the protein earthquake and the structure You would have the entire deep protein going through a major conformational change here And then it would release another protein and everything and I Probably half the proteins on this picture. We still don't really know the details of what they look like or how they bind We do have structures of these complexes now since 2011 and branco milchia But some of the most important research is still understanding what happening in all these interfaces. What if you have mutations in these interfaces? A great another great exemplary is not so much gpcrs, but signaling in general are viruses What if there is a specific virus? How do viruses work? So viruses they infect your body, right But viruses are also proteins. Remember the tobacco mosaic virus. So that's a code full of proteins So now you would like to identify and get something to bind to a virus But the virus is also really good at mutating So we're basically trying to create an ad we're trying to create something that should recognize something that is changing all the time And that's another challenge Can we then identify the specific things where it's difficult for the protein to change simply because they're most sensitive or something So can we get a very targeted fashion go after the parts of the molecule that are most important and they're likely most difficult for the Well protein whether it's a virus or another important molecule where it's most likely to have an effect either on the function Or at least in the sense that the virus can't change it And then in principle if we understand that we should be able to tailor the signaling Either destroy the signaling if you have too much signaling I might just want to shut off shut it off or if you have too little signaling Maybe I want to amplify it just five or ten percent to create a little bit more signaling I'm going to talk a little bit about that next week when I talk about protein design because that The goal is of course not to design proteins, right? But the goal is that I would like to have a very Fine dial so that I can start to change the biological process and make it Amplify it a little bit or dampen it a little bit So there are dozens of papers published on this both in simulations and Experiments every year now that we try to go after specific residues or tailor make new molecules that would have a specific effect And Ron draw in some of these papers they did they even they even they were man They managed to show not just how things bound, but as I already hinted in that simulation They show that the entire molecules moves from the active to an intermediate to an inactive state Based on the binding of this molecule. So now we no longer just talking about binding The process of binding is called affinity And the affinity just describes how how strong do the molecules bind to each other But this is what you in biology will call efficacy. How efficient is it? Binding to the molecule is just the first step, right? Does it have the effect that I hope it would have? Remember the one of the first slide I showed today when you talked about the agonist Just binding doesn't mean that you're an agonist The reason why this would be an agonist is that actually that it's Binding of the small ligands causes the protein to move from the active to an intermediate to say an inactive state Or in this case, it would be an antagonist, of course And this is not quite science fiction anymore because today there are dozens of examples of this in the literature Where people have been able to use computer simulations to not just simulate how things bind but As a consequence of the binding, how is the entire protein changing the shape and how is that having a functional rule? I Actually, I might let you go 10 minutes early today. I have a couple of more minutes. Um, why am I spending time going through this? This is not science fiction anymore. This is super important Gpcr drugs in particular two years ago when I gave this course this was published right in the middle of the course Uh, and astellas they bought a drug company for some 800 million euros So everybody is super excited about Well gpcrs in general, but also specific methods that make it possible to create custom drugs Because it makes you be that makes you able to go after in particular lifestyle diseases and everything We have a little bit of time so there are two things I want to bring up. Um, I spoke a lot about profit here and everything. Is this bad? How bad is big pharma? I know I know I did It is partly but I think that's the thing I told you that It is certainly a moral question. I'm I don't want to be the person protecting big pharma necessarily But on the other hand as part of my job, I go around and so first I have no commercial interest in this myself But we talked to quite a few of them The one thing that fascinates me is that all these companies they're filled with people who are passionate about curing disease Uh, and the the thing that tends to motivate them the most is talk to patients and talk to people that they realize their work is actually curing ill people And then the grand scheme of things The first that does not necessarily make these companies good. And if you look at some of the companies By far we talk they like to talk about how expensive it is to do the research and everything Um, and they spend almost as much money on research as they do on pr and sales So of course that they're not necessarily noble companies. These are coming on the other hand That's probably true for Volvo and the other company. They also spend a lot of money on sales And in the grand scheme of things suddenly if you start to compare that I think sweden has a proud history of say sob princess doing great fighter jets and everything It's proud swedish engineering tradition You could even argue from a security policy that it's important and everything I I don't see any moral problems of making fighter jets necessarily And suddenly in that perspective, but wait a second if we don't have any problems with making fighter jets and making money from that because we're good at it Making money from curing ill people because we're good at it Sure, it's a political problem of people how people should get access to the drugs and everything But if you're really talented at curing ill people, this shouldn't a good surgeon have a good salary Nobody's want to become a surgeon and I think in the grand scheme of things It would be worse if we don't have good surgeons So I would just like to Maybe modulate that statement. It's easy to think about big farming bad and there's there are tons of profits here They're probably not investing as much in basics research as they should But the main reason is that this is getting very very expensive and we as taxpayers are the ones constantly coming up with new ways And every time there's a mistake in a drug We are happy to sue them and we argue that there are now 10 different more tests that they need to do It will take them another three years Yeah, but they only had nine years to make a profit We just removed a third of the time where they can make a profit of the drug So that means they will not have to increase the price by 50 percent And then we think that's the company's fault And in a way it is or we can say maybe it's our fault That maybe the criteria we have for new drugs are so extreme and we're not willing to fund the research with taxpayers Or maybe I don't think we should So at the point that we are part of this equation, it's not necessarily AstraZeneca or any other company being evil The interesting thing the second interesting thing that's happening is that here I mostly spoke about small compounds The very big thing is happening is protein drugs. And that's what I'm going to speak about next week So proteins are super complicated because suddenly we have very flexible molecules So all these things about entropy and building a small suddenly doesn't quite hold anymore We're going to try to fold proteins in a way that makes it possible to create custom effects And that's super cool because if you if you think that these things are specific imagine how specific all these protein falls We've been looking at it. There are there are so many things we could do with proteins But we are playing with some Fairly big powers here to use suddenly starting to interact with antibodies and everything in the immune system So there are some pretty catastrophic failures too And I will I'm not going to spend a full lecture on that next week But there are some interesting lessons to talk about here The final thing I want to mention here is Given your profile one of the biggest trend That I'm actually quite happy to see is AI and machine learning going after not just disease but health So anything that has to do with differences fitness trackers and everything that's the said that's a low hanging fruit But there is an increasing awareness of trying to cure disease before it happens And this far it has mostly been based on health exercise and everything But what if we can find deviations in your genome that you are likely to get a disease in 20 years And maybe start taking preventive measures, you know what based on your genetic profile you should probably eat less saturated fat We all should but But there are going to be some minor deviation You could even it could be worse for the general person It's not so bad with a little bit saturated fat But with with your specific profile it's catastrophic You need to cut this down and do it now because you have 20 years If you do it now you're not going to have any problems with this That's going to be way cheaper to address And it's not going to be particularly fun to eat not to eat all those fats when you're 20s But it's probably more fun than getting the heart attack when you're 40 And there are a bunch of diseases like that One good example of the bioinformatics is Birken. Did Lucy talk about it? breast cancers So there are some Angelina Jolie, for instance, he has mastectomy where she removed her breast because if you have this gene It's pretty much between 85 to 100 probability that you will develop breast cancer before you're 40 And if you get these breast cancers, you die. They're exceptionally aggressive But of course because we can identify this genetic test today You can prevent it as a preventive measure remove the breasts And then you will actually not get it And if you think that's unfair it can actually It's less common But occasionally it happens to males too and if you're males, it's pretty much a guaranteed death because there's one thing males don't do They don't check their breasts So it's some small sign of fairness in the world, I guess I I'm going to leave you 10 minutes early today, but there are two things think about this And then there's usually one or two questions about this because it's important But I think it's a beautiful carryover between super advanced physics machine learning computer science But also health So the two big topics is Describe this modern drug discovery the pipeline how it happens the different stages You need to be aware of it and be aware On the one and how it is coupled to physics, but also how it is not coupled to physics Some of these things we are still doing better with trial and error Rather we're not doing it better with trial and error But the point is that we do it faster and if you can do something a million times faster Sometimes it's worth throwing physics out the window too And the second part is that we should have a little bit of an idea about this deep voting coupled receptors Why they are important and how they carry out their function by binding a ligand externally Changing the conformation and then creating a signal on the inside of the cell And with that I would