A THREE DAY HANDS ON WORKSHOP ON AI IN DRUG DISCOVERY & DEVELOPMENT

[00:00:10]Okay. Um so now we are going to talk about the another database which is uh called string. Okay. So if we do the network analysis enrichment interpretation and also we are going to talk about what are the common mistakes that the people do. Okay. And because we know that a gene a protein do not work on its own. Okay.

[00:00:41]It works together with some other proteins or some other genes. Okay. Uh that may be transformed into a disease.

[00:00:50]So in a drug discovery what happens is some people say that I'm working on this particular protein. Of course you can inhibit it or you can activate it.

[00:01:02]Suppose if you do not know that whether how this protein is interacting with other proteins.

[00:01:10]For example, you take the protein X. It is also associated with A, B and C.

[00:01:19]If you inhibit the protein A that's it will also inhibit all these three proteins and that's why it may cause uh some kind of a side effects or adverse events. Okay. So that's why it is called as the network analysis. So you need to use your uh knowledge to make a decision whether if I work on this particular protein. Okay. So what are the possible side effects? Okay. Or adverse events.

[00:01:53]Those things actually you can get from this string database. Okay. Okay.

[00:01:59]Strings help researchers move from a single target to a pathway informal drug discovery hypothesis by visualizing known and predicted protein associations.

[00:02:12]Okay. Now from this one suppose can I can we say that since you say that or can I work can I work can I put a title that EG FR my hypothesis is so and so compound is going to work on EG FR new that is not acceptable because it is telling only a a partial show you in yesterday we have seen that this EFR they have a three different pathways okay if you look at the slides so now you need to say which pathway specific pathway okay that one is okay and why this matters is that targets rarely act alone that's why I already told you cancer signaling is a network based of course if anyone looks at the cancer one really they get lots of break resistance may arise through bypass properties.

[00:03:16]Okay.

[00:03:18]And here in general what happens is in any network there are two things.

[00:03:28]One thing is called as upstream regulators then the other thing is called downstream mediators.

[00:03:37]These two are the terminologies that the people use in drug discovery. Okay.

[00:03:43]Upstream modulators means which is in the network pathway which is at the top level. So when you work your your compound works on this am right that is called upstream.

[00:03:56]They are called as a regulators because they are going to regulate all these properties. Then the other one below this one there will be some other genes or protein they are the ones that they produce response they will call it as downream mediators.

[00:04:12]Okay. So which one is good whether we should stop here or we should work at the bottom one. Okay. Suppose suppose if you work at the bottom one. Okay. Let us say x is the a b c. So you are working on A then still the disease may progress from B or C still. All right.

[00:04:38]So that is the problem that means your compound may not effectively cure the disease. Okay.

[00:04:46]Then but if you upstream one if you close down if you close down definitely A will stop B will also stop C will also stop. But in that one C is the one that is very important for us. C activation must be there. So if the C if you inhibit C also will stop that means we'll get some kind of a side effect.

[00:05:09]That is the reason why you see the nausea, vomiting, diarrhea all other things, muscle weakness everything that you will get because all these are found to be very effective.

[00:05:20]That's why people say that this is a breakthrough medicine.

[00:05:25]Why they call it is a breakthrough medicine?

[00:05:29]It is working on the upstream regulator.

[00:05:33]That's why it is able to do this one.

[00:05:34]This disease is going to be subsided.

[00:05:38]Then all other things are the people who take those medicines will suffer maybe with diarrhea, with nausea, vomiting, okay, headache and all these things. But what these companies say that oh all these are manageable forget about it. Okay. But they will sell it as a blockbuster drug. Blockbuster drug.

[00:06:00]Okay. Then because they say that because of nausea no one will die because of the vomiting no one will die. So that's what they say. But the patients they don't want. Okay. That's why in a cancer treatment even though it is progressing so much but many people are not willing to take chemotherapy because I know many people they say that yes I'm suffering with the cancer okay because if I take chemotherapy I lose my status in the society. So if I work if I leave for two years that's fine for me. Okay. and uh rather than taking this chemotherapy and telling to the entire public saying that I am a cancer patient okay the people don't want that social stigma is very very important okay and resistance may arise through bypass combination strategies need pathway okay this is the very important point now you look at the point am I right x A B C already I told you if you give it X what is the disadvantage and if you inhibit only one of the pathways A what would be the disadvantage now yes upstream regulator is good but it's also dangerous that's why you now you need to look at A B C C is good you don't want to this one but A B you want to inhibit then you need to use combination okay that's why we cannot even simply think that to treat a disease only one single medicine works. It may not. It may not. So whether in our drug discovery proposal whether we need to discover more than one medicine to treat disease completely for that we need to know this network analysis. That's where it helps. That's why you will see the combination strategies. Okay. Efr interacts with this family of proteins and these are adapter proteins activated MAP signaling then just now I told you that there is a mechanism that PK3/ AKT support survival okay network context explains response and resistance okay here use string to ask better drug discovery questions but not clearly to generate a colorful network image okay of course when you do this one it comes with a very colorful colorful image Sir. Okay. String is a hypothesis generation tool for pathway informal discovery decisions. Okay. Even though many people talk about this one, you know, still now the biggest challenge is inflammation.

[00:08:43]Inflammation causes so many things.

[00:08:46]Everywhere there is an inflammation.

[00:08:48]Until now there is no single drug. Okay, that works. That's why if you go to a doctor, he will prescribe all the medicines. that they will also prescribe parasol or something else.

[00:09:00]There is one next.

[00:09:03]Okay.

[00:09:05]This one here again you look at this validated target. This is the network content which proteins communicated with this one and pathway. Then you need to look at the pro target class which bypass mechanism matter. That means this is where target means whether you require a combination therapy or a monotherapy or what is the next step okay structure bioactivity and candidate design.

[00:09:31]So what it will do is because here this is the key message for each and every database or whatever the thing that I'm talking I'm talking about what is the thing that is good and what is the thing that is not so good. Okay. What does it mean? That means every database or every AI tool has one positive and also has negative thing. Okay. So you need to be aware of the negative things and you should encach the positive things that are associated with it. Okay. So which pathways are over represented? Which interactions have a stronger evidence?

[00:10:10]Which core target hypothesis are plausible? Of course if you do the practice that you will get it. Okay.

[00:10:16]String does not directly pro direct physical binding for every edge. It won't tell you and cancer cell specific interaction activity also it won't tell you because you are talking about EGFR.

[00:10:28]EGFR may be present in any also may be present in other diseases. Okay. But this network analysis won't tell you only about cancer. Okay.

[00:10:37]And drug it won't tell you whether drug is going to be effective or not and correctness of every predicted relationship. Again you have to go back and check. Okay. And best mindset is use string to generate a biological map then validate with literature and vomit data and experiments. Okay. There's a lot of a network is useful only when it changes your biological reasoning for next experiment. Okay. Next.

[00:11:07]Now here what are the four output that you will get is you will get a protein protein association network and you also get a conference scores for edges and you will see what is that edge and evidence channels and direct and indirect functional associations.

[00:11:23]So here function means the one that we see. Okay. So direct means when the receptor is activated automatically it produces a biological response that is called as a direct.

[00:11:39]Indirect means when you uh alter that one then it will go through some other protein then only it will produce that is indirect ones. Okay. And indirect ones is bit difficult. Okay. And it is called network clustering and functional enrichment analysis. Um this is a completely different technique. Okay.

[00:12:02]And here you can say that every experimental data curated pathway co expression G neighborhood text mining and also we can also do the computational predictions. Okay.

[00:12:13]Whatever you output you get you need to do this. Okay. And where it helps is it identifies the network context helps explain the resistance and suggest code targets. Okay. It also supports mechanism figures and whatever the figures that he gives even it's very difficult for us to even that itself is an AI tool. Okay, a tool is already embedded in the string. Okay, because in the gene card you have seen one human skeleton in which there are red dots. It is it's not a single static image. Okay, so um now disco value and supports mechanism and prioritizes follow string integrates multiple evidence.

[00:13:00]Therefore, evidence instruction is essential. Next question.

[00:13:05]Now this is the beginner question. What you need to do? What proteins are connected to EFR? Previously you checked for only ECF. Now you have seen that what are the proteins that are connected with EFR because in drug story you talk about the selectivity. You talk about the toxicity and all other things. All right. So for you to understand that one you need to know that which other proteins they are associated with it. So what are the nodes and edges? Which pathway names appear? Can I export the network image? Yes, definitely you can explore. Okay. And drug discovery question is what interactions explain EGFR driven growth especially in the case of cancer. Then what are the pathways and which partners are possible and what interactions deserve validations? Okay. Uh so here I want to say this one. This is one of the common mistakes what people do in an academic research.

[00:14:03]when you say uh I'm taking a simple example of cancer okay they take particular thing they ask the they culture the cells in a plate and they will put it will be killed and they say that they have anti-cancer effect it's not correct It's not at all because anti-cancer there are three different terminologies that the people use cytotoxicity, anti-cancer, anti-tumor.

[00:14:47]Okay. Cytotoxicity means whatever we do in in vitro that is called cytotoxicity.

[00:14:55]Strictly speaking, we cannot use the word anti-cancer for that.

[00:15:00]Anti-cancer means if you use in animal model, it shows the effect and we should call it as an anti-cancer.

[00:15:08]Okay. Then if the compound shows the activity in a human, it is called as an anti-tumor.

[00:15:16]Okay. And many people do not know, even I also do not know when I started my research. Okay. So actually I started my research when I moved to Malaysia on cancer research because I was so passionate as everyone thinks that the cancer research is the great greatest like that I also started. Okay. Then I did I got lots of compounds everything.

[00:15:38]That time the system came with a 3D culture 3D culture. Okay. That is the most appropriate thing to check. So I used to say so great about it and uh the people also used to think that is a great cancer research work those days. Okay.

[00:15:56]Then I also thought that I'm doing a great job. Then I went to I got one person in UK and those wife and husband they started a contract research company and they said that now we come up with a new thing because in 2D cells what happens is the cells are already clustered you know they're competing with each other and I know the lots of tips and tricks even the compound is not anti-cancer I can show the experimental proof it is anti-cancer I can lots of tips and tricks are there I don't put the nutrients that's all the cells will die that's all the cells die because of my drug. Yeah, lots of things can be done.

[00:16:32]Okay, but in reality that is not there.

[00:16:34]Am I right? And actually you you are sticking onto it. You are sticking onto something. It is called as radical cells. Okay, they're sticking onto it.

[00:16:43]Okay. So then they come up with a 3D culture. When I put the compound into 3D culture, they did and they said that sorry none of your compounds are active.

[00:16:54]All my seven years work has gone. Then I cut up with the answer. There is no way that I can do it. Okay. Uh then I cut up with it and listen because why I'm telling all these things is okay. Of course uh when you are young, yes, you can use the words for everything you can use the word antic. But when you are growing up actually you need to understand what are the differences.

[00:17:19]Okay. So uh yeah that's why interaction explains what branches where I'm talking EGFR >> discovery question okay which EFR associated proteins and enriched pathways support a combination strategy for EFRK resistance in CLC you see previously what I said only EFR that's all right now I'm going little bit advanc you don't consider only not only EFR you also should look at other proteins okay which are closely associated with EFR because your target was not to your your end strategy was not to synthesize very fantastic molecule that works only on ECF okay your strategy your aim is you want to treat the cancer and you know that EGFR do not act alone it also acts with other proteins so you need to put them together Okay. So that's what we that's why the stream database is going to come. Okay. This question guides settings interpretation and reporting.

[00:18:26]Okay. Do not start with the network.

[00:18:28]Start with the decision the network should support. Next.

[00:18:34]Okay. And a reproducible approach.

[00:18:37]Actually when you see this one I always use the word reproducible because it has to be this one. Okay. Then set to the correct protein. So you have to use the protein name, gene symbol or accession.

[00:18:51]I think my colleagues say that you should use uni plot number. Okay. And and select organism carefully. Sometimes the people don't select the homo sapiens that is human. Okay. Start with the default network. Don't make any changes.

[00:19:05]Then afterwards if you want to play around you can adjust the settings. Then these are the ke and go and react. These are the very powerful pathways. Very very very very powerful pathways. Okay.

[00:19:20]And uh which is very difficult for me which is difficult for anyone to understand. Uh and actually these data these ones I will ask my team uh to explain to me. Okay. It's very difficult for me to understand this. Then you have to export and document it. Okay.

[00:19:37]Uh run the same EF test twice first with the medium confidence then with the high confidence. Of course you can check it.

[00:19:43]This demonstrates that settings are not cosmetic. They directly affect scientific interpretation. Okay.

[00:19:50]Okay. Reproducibility requires regarding both the query and network settings.

[00:19:55]Next.

[00:19:56]Now this is what you are trying to get.

[00:19:59]These are the different different images very beautiful images and all those things will occur. Okay. because actually these kind of a beautiful images will also help uh to succeed uh in getting the publications. Okay. Uh and also in successfully obtaining the research grants but however whatever the image that you put you must be in a position to explain. Okay. Then the problem will start. Okay. Note means a protein or gene product. Okay.

[00:20:33]Previously.

[00:20:35]Yeah, edgy means association between the proteins. Okay, when you put more edges then you say more connected network and if the line is thicker that means they are closely associated and the clusters that means when you see them together then they say that related functions are pathways. Okay, nowadays the people are talking about AIA and all those things but these databases are there very long time back. Okay, maybe 10 10 15 years back itself is there but those days the people never used the word called artificial intelligence but nowadays the artificial intelligence became a fancy term. Okay. But the all these are really there in the process. Okay. Now the scientific interpretation asked whether each relationship is biologically plausible. Again how you check that one is you look at the literature and it is supported by a strong evidence and it is relevant to the cancer and it should be related to the proliferation survival or resistance. Okay. In cancer research we need to know the differences between these three terminologies proliferation survival or resistance. And here also most of the people who do the research okay in cancer this is the common mistake that they do what they do is they they put the cells okay they allow the cells to grow and in another compartment they will put a drug it will allow to grow maybe 24 hours 48 hours or 78 hours it depends upon that afterwards they will they will compare these two and they say that the cell count here is less compared to the untreatment.

[00:22:18]But from that how can we say that it is killing the cells? Actually it is preventing the growth of the cells. It is preventing the growth of the cells.

[00:22:28]Okay. That is called as the proliferation.

[00:22:31]Okay. That's why strictly speaking it is called as an anti-prololiferative effect of the compound.

[00:22:38]Am I right? Because in the treatment what we want we want the compound to kill the cells. It's not to prevent the growth of the cells. All right. Okay.

[00:22:48]Okay. That is also required. It also should kill. And most of the experimental protocols that they do in anti-cancer research, they do only anti proliferative action. They don't do other things. Of course, that itself is a big topic. The another one is a survival. Survival means you put the trade component.

[00:23:11]We are not telling why 30% is surviving remaining 30% is really surviving and no one talks about it. Then we talk about the resistance. Okay. So when you treat treat okay this also I will tell you because cancer cells is one of the uh difficult to procure once. Okay. And most of the laboratories what they do is whatever the cell that is available they will get it and they do it and they do not know what is pass number. There is something called a passage number. Okay.

[00:23:44]A passage number means when I take out from the uh supplier I will put into the nitrogen one that is the first pas then afterwards I do a subculturing that pas 2 pas 3 okay and strictly speaking from pas 4 to the pas 10 only we should use and apart from that we should not use and most of the most of the papers that I most of the papers most of the students who present the anti-cancer affected. The first question that I ask is could you please tell me what is the pas number of the cells that you use and if they do not answer it then I say uh okay you just proceed but your results are not reliable and okay those are the things that we need to be quick and because why the pas number 10 this one means because it develops the resistance that's why number two 42 pas number 10 is the most important Then you may ask why why can't we use process number one and process number two. The reason for that is actually when the cells are shipped they are shipped under ice. Okay. At usually at minus 80 and then afterwards you are putting into the nitrogen cylinder.

[00:25:01]Okay. Then all these things you need to slowly bring it down to the room temperature. That's why it requires pass one pass two pass three and pass three.

[00:25:11]Yeah.

[00:25:13]So the network is a map of thinking. It is not the final final biological proof.

[00:25:20]Next one.

[00:25:22]Yeah.

[00:25:25]No next one. Uh okay. Here this is what I say. Am I right? Because this kind of a word confidence you will find in lots of the databases. Okay. More medium confidence, high confidence and very high confidence. Okay. Here you can see that when you put a medium you will get a more nodes initial exploration and brainstorming. Okay, just look at the picture and think but don't do anything with it. And high confidence when you are going high your notes your this thing will appear the image will become lesser numbers. All right. So this is useful for the reporting. Okay. Then very high confidence uh you may fail. Okay. Uh then more interactors. Okay. Then you can see more things and fewer interactors you may miss out something and even reveals why an object exist. But again you need to look at the scientific interpretation.

[00:26:24]Okay.

[00:26:26]Because here again towards the end in drug discovery research you are the boss.

[00:26:33]Okay? Don't give that role to any research assistant.

[00:26:39]You are the boss. You make a decision.

[00:26:42]You make a decision whether you want 0.7 or 0.6 or 0.5. Okay, that's very important. Okay, higher confidence improves reliability but may reduce biological. Okay, the next one.

[00:26:55]So what enrichment ask is are the proteins in my network over represented in a specific biological functions or pathways compared to with what would be expected by a chance. Okay, that means uh let us say if you take out the inflammatory pathway, it is also there in cancer, it is also there in diabetes, it is also there in uh other diseases.

[00:27:17]Okay, actually in enrichment actually it will tell you which disease it is more predominant. It is more predominant.

[00:27:27]For example, I can tell you how many of you know that metformin is also useful to reduce the weight.

[00:27:38]How many of you know? Yeah. It is one of the one of the best drugs that the people think that uh it is good for the weight loss. Okay. Yeah. So why means if you look at that entire network actually you will understand otherwise people think that metformin is only okay. So of course now recently because of nitrosamines and all those things there has a problem.

[00:28:05]So the why I give that example is because the way that metformin works am I right? the highest enrichment is there in the diabetes and the second enrichment is something related to obesity that's why it is also helpful okay so that that is that's why I I I tell that example okay so what to look for you have to look for I told you about the pathway EFR you have to be signaling MAP pathway P3D okay there are so many pathways and receptor thyroid activity okay use enrichment to write a concise pathway rational That means okay when we started this thing we started with the cancer okay lung cancer cancer then we narrow it down to lung cancer from lung cancer to we narrow it down to the non small cell lung cancer from there we come to the EFR from the EFR we come to the EFR mutant from the EFR mutant we come up with a pathway so this is how you are narrowing down the uh discovery You know, enrichment converts network graphics into mechanistic drug discovery argument. Next one.

[00:29:19]Okay. Uh and what to search and how to structure it? You need to look at the target context. Okay. And pathway mode select multiple proteins. So this is focused once and candidate co targets and expand interactors inspect evidence.

[00:29:36]Okay. and omix interpretation input a gene list of from transcrytoics or proteomics. Okay, this is the one. All right. Uh here I want to take something how how this one how I use this network. Okay. To to perform the experiment. Am I right? So in the easy network when you do you will get lots of pathways. Am I right? And I will check let us say X A B C. Am I right? So my experimental protocol, my first experimental protocol is on X.

[00:30:15]Okay, I will check it. So whether it works, it's preminently pass. Then I check this A that my second experiment is A. Okay.

[00:30:26]Then I do for B, I also do for C. Then afterwards I go look at whether anti-inflammatory effect or not. This is called as a complete discovery proposal, discovery experimental protocol. You cannot say that some people will work only on this one. I discover something or some people will do at the bottom one. They will say this one. So compute complete discovery means you are coming from the upstream regulator downstream mediator and the effect and for each and every experiment there are two what is that called as there are two experimental protocols. One is predictory or informative. The another one is a confirmatory compound. Okay, that is always important. You cannot confirm the activity of the compound by a single experiment. Okay, uh because why why I'm stressing so much is in my initial per my research proposals were rejected so many times then when then I inspect then I know what mistake that I have important. Okay. and truck mechanism figure mechanistic illustration for reports and evidence grounded AI use.

[00:31:39]Okay, this one is whatever you get actually you can use any AI. Okay, do the things.

[00:31:50]Look at the database.

[00:31:52]Put that input into the AI.

[00:31:55]Then you look at the output again. You confirm it.

[00:32:00]Don't work in other way. That means never ever start your research with your research assistant.

[00:32:09]Okay? first to the database then afterwards you can give to AI then you will get the fantastic fantastic proposals. Okay. The best strength such depends on whether you are exporting a target a portray or anomix. Next okay practical habits that improve the scientific quality always concern the identity. Okay.

[00:32:36]for easy do not accidentally analyze a non-human protein. Okay. Inspect evidence.

[00:32:43]Okay. Whether support comes from the experiments, curated databases, co-expression, prediction or text mining. Okay. And we should always believe the experimental data only.

[00:32:54]Okay. Not others. And keep it readable.

[00:32:57]Okay. Use high confidence. Limit interactions. Export the figure. Okay.

[00:33:02]Good string analysis selective, documented and biologically interpreted.

[00:33:07]Next one.

[00:33:10]Yeah. Avoiding misleading network conclusions. And these are the common mistakes what the people do is this one.

[00:33:18]Okay. And and these are the better practices and these are the problems that are associated with this one. Okay. And the biggest mistake is treating string output as proof rather than evidence to interpret because nowadays what the people are saying is whatever it come should be right. That means we are we are thinking that the other researcher who developed this string is much better than us. Okay. Next one.

[00:33:47]Okay. And here also very similar string show that EFR interacts with many proteins. That is that's not correct.

[00:33:55]Actually say that a string network centered on a human ECF form. Okay. The thing is here you are saying that which database you are talking about the human. Okay. Was generated using a high confidence interaction threshold. The network was enriched with ERP map and EI3K signaling supporting EFR as a pathway connected driver of proliferation and survival. Okay. This is what you are going to tell. Okay.

[00:34:25]Then report the method result interpretation and the limitation to the blood. Okay. Always you need to learn the limitations. Okay. Next.

[00:34:35]Okay.

[00:34:37]So core target should be chosen to solve a defined resistance or pathway bypass problems. I told you that simply our our our discovery is our intention is not to discover a fantastic molecule against the target. or this one is to treat the disease. Okay. So these are the potential easier for relevant partners.

[00:35:00]All these are the pathways. Okay. How to use this type of Okay. Uh do not immediately claim a combination therapy and instead ask is the code target active in my disease model. Is there biomarker evidence? Is the combination tolerable? Can I test it experimentally?

[00:35:16]This is the very important thing.

[00:35:18]everything whatever you do and you should be able to do it experimentally.

[00:35:23]Okay, next one.

[00:35:25]Yeah, these are the take more messes.

[00:35:27]Okay, string places a target inside a biological network. Interactions can be physical enrichment can be done. The next step is literature or experimental validation. Okay. Uh do not use string to decorate a project. Okay, please.

[00:35:42]Okay. Next one is I think this is the last one. Last one.

[00:35:49]>> Yes sir.

[00:35:50]>> Okay. Uh can you open that one? What is >> Okay. What I do is Okay. Just go through this one.

[00:36:24]Yeah, this is the string database. Okay, there is a step-by-step procedure is there. Okay, next.

[00:36:32]Do only these things. Again, I am saying that if I want to explain all these things, it is going to take a very very long time. Just read do it slowly. Okay.

[00:36:43]Do it day by day. Sorry. Uh on this one.

[00:36:47]Okay. Next one.

[00:36:50]Next please.

[00:36:52]Yeah. You will get a very uh in general in high impact factor publications.

[00:36:59]Actually you can see this kind of a network. Okay. And actually it improves uh the chances of the paper getting published. Okay. especially those who are writing a review articles these kind of images really help okay and nowadays uh since I'm talking about the images the success of your project or the success of your uh paper acceptance it entirely depends upon the image okay so usually my my strategy is before we write any paper we ask my team to generate three images first then if I'm satisfied with those three images then only we go and write starting writing a paper we don't write a paper first we write because without images no one is going to accept okay and why they don't accept whether it is right or wrong of course uh I can talk about it for about half an hour why the trend is happening okay these are the things and here also So they say that use more options do this next. Yeah.

[00:38:11]Seven 8 next. Yeah. You will get so much this one and next one. Yeah. And just you record your these things here. Okay.

[00:38:22]So that is about uh that is about uh this one. Of course these are the some the some of the tips that we do. Okay.

[00:38:34]So uh can you open strings database?

[00:39:03]Yeah.

[00:39:05]Yes.

[00:39:14]Okay. Uh here. Okay. You can go to search.

[00:39:19]Actually uh big people what they do is here you can put easier for >> actually here you can see that spring chart new model this is the right this is the yeah so in that one you can you can search Yeah, here you can play around with here. If you see this one, am I right?

[00:40:12]Uh but can you It's good actually. Of course, different colors indicate different things. Okay. Thickness indicates lots of things actually. Okay.

[00:40:24]Now, okay. This is what we can use it.

[00:40:27]Now, if you see this one, you're getting confused. Am I right? Okay. What you can do is you can copy this image, put it into the generative.

[00:40:39]Can you please explain this image in a layman text and it will explain >> that's where you can use your research assistant and from that once you understand that hey what you are talking about this CBL I did not understand what is it CV it will tell even if you say hey you are telling me in English can you tell me in Telugu that also will tell or it also tell no I don't have a habit of reading can you speak for me it will speak for you that's where the AI will help that's how we should use AI but you should not use AI to generate this map that's not responsible use okay so then go down little bit here you can see that of course leg settings analysis exports clusters there's a lot of things that you can do node colors what are the object Can you what are the known interactions these things and actually you know interpreting this data itself is the biggest thing this is the here you can see the conference pro usually we go 0 to 0.7 even after you do that one also okay you put this one entire thing you can copy and put it into generative AI can you oh now I basing on this one I want to develop a hypothesis can you help me to develop a hypothesis for it will develop. Okay. Once it developer, never ever blindly accept whatever AI says.

[00:42:19]Whatever AI says, okay, it can bring you to the very high level or it can bring you down to the very level. Okay?

[00:42:27]Please, that's why don't put so much hopes on AI. Of course when I talk about alpha fold I will tell you why this is. Okay. So this is about this one. Okay. I think what we do is we try to do one more lecture alpha 4 afterwards. Is it okay?

[00:42:53]Okay. It seems uh the participants are good. At least uh I don't see anyone sleeping. Okay. That is good thing. And uh the second one is uh not making a noise. Okay.

[00:43:09]And of course sometimes while opening eyes also we can sleep.

[00:43:15]Okay. Uh that is called as a brain sleeping. Usually it happens. Okay. But it's okay. Okay. Now here this alpha fold this is the one thing actually that is going to take away the people are getting worried in a drug discovery research. It is going to take away lots of lots of people out of the jobs in drug discovery.

[00:43:47]Okay. So no one knows whether it is good uh whether is a good or whether it is not good. The people who are fancy about this fancy term artificial intelligence they're appreciating.

[00:44:04]But the people who are really struggled worked hard in research and developed some kind of an expertise spend about 20 to 30 years and they now they're thinking that oh they are going to become users.

[00:44:20]Okay. And this alpha 4 the uh what is it who has introduced this? Okay. He got the Nobel laurate prize. Okay. And it is a biggest surprise to the entire scientific community in the world.

[00:44:44]Okay. How come a a particular scientist is being given a noble or a prize? Okay. that does not have any experimental evidence until now all the noble okay especially in this one only for the experimental work okay but this is the guy there lots of things can be learned from this guy so there are both pros and cons people are talking a lot about it but the message that I got from here from this guy is basically he's a biochemist He knows the subject.

[00:45:30]Okay. Then he's making use of the research assistant.

[00:45:34]That's why he wanted. That's why he became more successful.

[00:45:39]Okay. What I mean is you are a pharmacy or you are doing a biological sciences research that is your core expertise.

[00:45:51]Try to master it first then go for the AI how it is going to help you don't waste your time to master AI okay try to develop this knowledge then whatever the whatever the things that you got this one okay then I also told one word that saying that why some people are of course this was happened only in my recent visit to UK when we are discussing actually uh you know in a designing molecule what we need we need a protein and we know we we need to know it's a 3D structure once we know the 3D structure then we need to see whether the our compound is fitting in or not that is entire about computational that's all nothing else right so to get this protein the people used to do X-ray crystalography Okay.

[00:46:52]Electron microscopy even they also use NMR. Okay. That they used to take a very long time. All right. And that is called as a structural biology. And that structural biology is one of the most important key concept in a drug discovery because if you do not know the structure of the protein there is no way that you can design the molecule. Am I right? So then when these people have said this one they said that we are going to lose all our jobs because this guy is trying to predict a lot. Okay. So that's why even uh even I'm struggling with one protein for my research work then why have I asked them they say that hey why can't you try with alcohol then I say oh why you are saying that one so what to do that is the reality am right so and this is one of the best examples where you need to say that that we need to stick to our grounds first okay don't try to jump one to one one to one and you know rolling stones gather no mass always okay that's okay now why this matters for the EF this one is protein structure is the bridge between target biology and mole what I told you alpha 4 DB gives a rapid access to the AI predicted protein structures and confidence metrics okay again I told you confidence is defined for drug discovery the value is not simply viewing a structure it is knowing what can and cannot be trusted okay then of course this one we are talking about can we use the predicted EFR structure safely?

[00:48:32]The key word is safely to understand domains, mutations and possible binding site context before moving to the experimental structures and legant data.

[00:48:43]Okay. And you see alpha is a powerful starting point but not a final drug design answer. Okay.

[00:48:54]uh here I want to tell that I think I have written my first AI paper I think during COVID time okay in a drug discovery today okay uh since we cannot go to the lab so uh all the labs are closed then my team asked me a anyway you won't uh keep point we have to do something okay anyway someone is talking about AI why can't we talk about AI what is this real use in drug discovery that is the first paper that we publish in the drug discovery today in 2019 or 2019. Okay. So at that point from that point until now everyone is saying that they do so much almost 8 years 9 years have gone. Can you tell me is there any product that came from the air which is there in the clinic?

[00:49:54]Not not at the clinical drugs. still at the FDA. No, until now there is no single drug there is no single medicine which is approved by US so far and uh because uh I met uh I met few of the I'm not getting the names who have started up this company uh startup companies of AI drug discovery I'm forgetting his name uh he is from the University a friendly product and uh uh I'm not uh I'm not getting the name of the company. So straight away he says so many things. Oh that that that so many drugs so many drugs reach into the clinical trials and all those things and later on no one knows what happened to him. Okay. So sorry uh not even safety even efficacy also it is not it is not possible even it is not passing phase one clinical trial or not not at not at this one. So here I will tell you okay why this is my only my my gut feeling is why AI is not working for practice story.

[00:51:21]Unfortunately it is because of the fake data that is available in the literature.

[00:51:28]Fake data available in the literature.

[00:51:32]Someone say that the anticancer of activity IC50 activity of this compound is 10 nanom. The another one say that 10 microar the another one say 100 nanom it doesn't know now that is the biggest problem now that is really really the biggest problem.

[00:51:52]Okay, of course later when I'm going to talk about keml then I will tell you you will see that for the same drug you will see the different values for the same target. Okay, that's why it is it is that is also put not able to that's why someone has asked me a if to be successful okay in deter what should we do one thing what we should do is we should remove all the literature that is existing in the world now I haven't edited it because all that is a fake data everything is a fake everything is a real even including patent is also fake okay when I'm saying that 70 to 80% of the things are fake.

[00:52:33]When I say fake, it may not be intentional. Okay? It may be an experimental error and all these things.

[00:52:39]That's why you know when I put myself, it will recognize oh this is a because but if I do do a cosmetic changes and go then it may not recognize me. Am I right? In the research also most of this thing is a cosmetic things because The goal is again the problem is writing a research paper and completing a PhD thesis or completing a M for thesis or anything MSE thesis. So uh there are lots of things like which I do not want to go into that but that is my personal view. It is nothing related to my university view.

[00:53:22]So and then okay uh alphabet is this one start with a scientific question and not with the website. This is what I have been telling so many times. Okay. So that it will impress upon next.

[00:53:40]Okay.

[00:53:42]Target validated sequence identified alpha model confidence check. Of course, if you are lucky, you can check the PDB because all the PDB is the experimental design. Experimental designs and design hypothesis. Of course, our my next lecture I'm going to talk about PDB. I will also tell you what happens.

[00:54:02]What alpha helps answer? What is the likely 3D fold of the protein? The important is likely.

[00:54:12]It is not saying that the absolute 3D structure. It says that how likely it is going to look like. Okay. Which domains are structured and which may be flexible. It will tell you. Okay. Is a region suitable for structural interpretation? Yes, it can tell you which residues or domain should be investigated further. Okay.

[00:54:35]It cannot tell whether legant will bind to it or not. It usually lacks B flux, B factors in PC specific context because I told you that in a string database you say that for one protein that protein will be active only in the presence of another protein.

[00:54:56]So you need to crystallize these two together. Am I right? Then only you can see that confirmation because suppose if you see take a one protein it is like this. Am I right? So it looks like this.

[00:55:08]When I'm putting another protein nearer to it, it may flip like this. The confirmation may change. The shape may change. When the shape changes, your drug may not fit in. So that one alpha cannot do. It does not replace PDB structures. It does not remove the need for experimental validation. Okay. But even without that also we got it. Next.

[00:55:32]So what the database provides is a precomputed protein structure predictions generated by using a code buttons okay uh and search by protein name 3D viewer confidence merits okay and here the good thing is it's a rapid you whatever the protein that you are working on actually you can use this okay uh here I want to tell my my own personal experience okay uh actually at this point of I'm working on one project uh in which I'm going to synthesize uh a new molecule working on candida multi-drug resistant candida ris okay until now there is only one drug molecule which is there in the market that is by fisa okay that is called manipics okay but still it is in the phase three clinical trial but it is being used for accelerated use okay so when I did that I started with a great enthusiasm that I think I'm working on a particular engine called GWT1. So then I said that that is the first time in my career why I cannot synthesize GWT1 engine okay because GWT1 is not at all there that protein no one knows how that protein is there. Then I made a bold step to see that whether I can synthesize the GWT1 protein itself on my own. Okay. uh so one by one means uh my research team but uh by taking the others help we did all the things we can we could able to get a GWT1 am I right then but we we could not able to get any form am I right so ultimately again alpha is one that helped me okay then go to the GWP1 predicted structure okay so of course whatever the compounds that I got only 20% showing activity. Remaining 80% is not showing activity. Okay. So in that way that means there is something in it in alpha fold but it is not always because uh without this alpha fold I might have not carried out that particular uh particular project because there is no protein structure that is so in that way it will helps okay that's why it is called as a precomputed protein structure. Okay. And all these things. Okay. And rapid structural starting point. Okay. Well, then when no experimental structure is available that is my my example just I told you that supports domain mapping.

[00:58:09]Okay. Helps plan which expanded structures are needed text useful for teaching structure function relationship in target. Next and this recommended such flow. Okay.

[00:58:23]Please this one.

[00:58:26]Among all these databases, uniro is uniro will give you a code.

[00:58:33]Okay. And that code you should use. That is the most ideal way of doing things.

[00:58:40]Okay. And check routine length and domain descriptions. Download and record alpha entry. Record the exact accession.

[00:58:48]Okay.

[00:58:49]uh and use identifiers carried forward from the gene cards or uni brought rather than only common names. That means all these are linked to one each other. Right? So search for this one family proteins same homo sapiens. Okay.

[00:59:06]Search by accession confirm by biology then only interpret. Next one.

[00:59:12]Yeah. Next previous.

[00:59:16]Yeah. Here you can see this is the P 000533.

[00:59:21]Okay, because in gene cards you say that EGR can be known by other names. So you need to know all those uh alias, right?

[00:59:31]Whereas if you use this one a single thing is sufficient because that one covers everything.

[00:59:38]Then if using gene symbol confirms organism and entry name check protein length download or record the exact accession. Okay. When did you downloaded and date because database interfaces and converge may work. Okay. And um yeah I next next.

[00:59:58]Okay. This protein identity 3D model viewer conference colors of course you can download. Okay. This PDA PDP file you can download. And once you download whatever you do is you put it to uh all the software that the first day yesterday he he showed about docking and all those things. Am I right? That one you can uh you can do it. Am I right?

[01:00:21]And do not download and use your model before checking identity confidence and basics. That's the key. Next one. Okay.

[01:00:30]Now this is thing very high often reliable for folded domain interpretation. 70 to not 90 is okay. 50 to 70 low less than 50. Okay.

[01:00:43]And P helps judge whether the relative positions or domains are reliable. Okay.

[01:00:48]Because you imagine that how it is predicting. How it is predicting?

[01:00:57]Can anyone guess how how this alpha 4 is predicting the structures?

[01:01:09]Okay, it's a pure pure chemistry.

[01:01:15]Protein is what? Amino acids.

[01:01:18]When two amino are next to each other okay depending upon the nature of the functional groups either it may ripple or it may attract or it may fold that's all because this is only 20 amino acids that those 20 amino acids depending upon how they arrange in a sequence it will commit 3D confirmation that's why if you look at the RFCPD there is more than 100,000 proteins you imagine This simple 20 amino acids are creating so many proteins. Okay. So that is the thing you need to know.

[01:01:55]Okay. Uh confidence source are not decoration.

[01:02:00]They define the level of trust. Of course if you go more than 90 is very high. Next one.

[01:02:09]Then step by step. This one is a beginner friendly. Okay. Search easy.

[01:02:14]Confirm the entry name. Identify extracellular transmembrane or and intracellular kinus regions. This is important because we know that in the gene cards we say that efr in a chromosomeal location it can present in any place all right in a cell. So we usually we need to look at the extracellular and of course I think for EGFR there is a PD structure check.

[01:02:41]Okay. EFR is a receptor alpha can explain overall domain organization but we need to always uh check the PDB.

[01:02:52]Okay. And of course we can also change this. Okay. Next. Next one.

[01:02:58]Okay. Uh and these are the practical applications. Okay. You can understand the target. You can develop a hypothesis.

[01:03:07]Of course, you can also write how the mutations interpret and also you can build the model and also you can also do this one. Okay. Structural hypothesis generation plus experimental verification. All this you need to do the experimentation.

[01:03:22]Do not use it blindly.

[01:03:24]That's very important. Claiming exact drug binding poses without legant B validation. talking into low confidence or highly flexible reasons. Okay.

[01:03:34]Ignoring alternative confirmations and making it uh conclusions. Next one.

[01:03:44]Yeah. These are the common some of the common mistakes that what the people do is okay. Uh yeah. Next one.

[01:03:56]And this one before opening the alpha field alpha 4 you have to define again the research question you have to identify the correct identifying the gene cards and you need to know what is its biological relevance. Okay and you check whether the experimental structure is available in the PDB or not. Okay, while you are viewing the model and you look at these colors and you look at avoid exact residue level claims in low confidence regions, okay, because if you open the protein structure for each and every region, it will tell you the confidence levels. Okay, and take screenshots. Okay, only after adding interpretation, not as a evidence alone.

[01:04:41]Then after viewing you compare with PDB structures click literature then record then use the model okay and combine alpha for PDB literature and bioation next then this is the minimum reporting checklist. Okay, all these things. Okay, name, gene symbol, everything. Okay, and example reporting sentence. Okay, this is what you need to say. The alpha fold DB model for human UCFR.

[01:05:17]Okay, you see within the bracket you need to put the uniro ID was inspected to understand overall domain architecture interpretation focused on structured regions with high PLDDT components. Legant binding conclusions were not drawn from alpha gold alone and were compared with inhibitor bone PDP structures wherever it is. So this is what you need to see. Okay. Why reporting matters means it improves the reproducibility. It prevents the war claiming helps reviewers shows scientific maturity. Okay. Next one.

[01:05:55]Yeah. Now you actually you can see that alpha code plus PDB there that what we are going to talk about we are going to talk about popc Swiss AB all these three are interrelated all are interrelated okay when it comes to the drug disco point so it will tell you the domain architecture these are the experimental structures this one will know the what are the reported compounds these are tell you that whatever the compound that you are thinking whether they are similar or not And these are the ones that they will tell you that they will tell you that uh how toxic it is. Okay. Then alpha for becomes more useful when connected to the orthogonal.

[01:06:41]Next one.

[01:06:45]Yeah, this one. Start with a defined target question. Use a reliable identifier. Confirm organism all these things. Okay. Of course, these are the workshops one. It is already there in the notes. Okay.

[01:06:58]Uh do not ask what structure can I give.

[01:07:01]Ask what decision will this structure support. Okay.

[01:07:07]Yeah. Next one.

[01:07:11]Yeah.

[01:07:12]Alpha 4B is a highly valuable browser base. Okay.

[01:07:17]should be integrated with gene cards top string prediction, interpretation, verification and experimental planning is strongest when used by scientists who understand its actually this particular statement is applicable for every uh AI tool. Okay, next. I think this is the one.

[01:07:41]>> So, uh this is the one and in alpha code actually we did not give uh what I say is uh in the workshop I put the alpha fold and PDP together. Okay. But I did not give a much emphasis on alpha fold. But I give a much emphasis on the PDB so that you will understand more science then later on you can explore more on alpha and alpha fold is the one we should really watch uh in near future because it may change the entire entire okay. Uh, thanks. That's the I think Okay.

[01:09:14]Of course, everyone says no. But I asked you if you realize that the one of the Nobel or uh Nobel Prize winners is uh uh that guy. Okay. Because uh to me that is the first instance where the AI is being respected by the scientists.

[01:09:50]Until now in this part of the world we see AI is very high. In other parts of the world usually the people don't accept it. Okay. and but alpha fold is the one where the people are talking about uh is going to be game changer. So here the take home message is now is uh now you can discover medicines for any topic that is the take home message with alpha 4 okay that means what I say is you discover But I'm not sure how many of them will be converted into medicines. There is still a big question. Okay. But it's okay in another 10 years or 15 years definitely it is going to be.

[01:10:43]So the take home message and also it is also one of the examples that you should be domain expert that take the help of AI.

[01:10:56]Okay. Don't make use of AI for your expertise. These are the key messages.

[01:11:24]Thank you sir for your elaborate session.

[01:11:27]Now I request all the participants both online and offline to proceed for the len session and the afternoon session will start at 1:30 a.m. and the live stream will be proceeded from 11. Thank you everyone for joining this session.

[01:11:41]The lambda stream.