 Hi, everybody. I'm Arman Aksoy. I'm a graduate student working with Chris Sander. It's Memorine Sloan-Ketrin Computational Biology Center. And before presenting more work, I would like to thank the organizers for giving me the opportunity to explain and present our recent work. And then in the next 15 minutes, I'll be basically talking about this project that we call study is a project that's named after the Roman poet, who first described the Achilles story. So as the name goes, as you can understand, I'll be talking about, in this, for this project, I'll be talking about individualized therapeutic vulnerabilities in cancer, so-called Achilles heels, and then how we predict them from the genomic profiles. But before, in order to give me, give you an understanding of what we call a vulnerability in the context of this project, I first have to give you a little bit of background of information about the biological processes. And then here, in this figure, in this really simple cartoon, I'm showing you a normal cell. Again, this is really oversimplified. And then you can see that within the cell, there's a metabolic reaction happening. The reaction is represented by this rectangular box. And then as you can see, there are inputs and outputs to this reaction. And these types of reaction happen all of the time. But cells are smart and they want to regulate these type of reactions. So within cells, we have these biological enzymes that regulate these type of reactions. And again, cells are smart. They don't rely on only a single copy of the enzyme, but they usually have multiple enzymes that catalyze the reaction. So in this case, we have two enzymes. You can have two or sometimes more than two. And then these type of enzymes are called isoenzymes in the biological context. So of course, this is what happens inside a normal cell, but we are interested in what happens in a cancer cell. And then on its way for a normal cell to become a cancer cell, there are lots of things happening, of course. And then there are many factors affecting this process. But the one that we are particularly interested in are the homozygous deletions that happen due to genomic instability of these cancer cells. And as you can see, I'm right now depicting a homozygous deletion just by chance. A homozygous deletion that just by chance affected one of the enzymes. And due to this, the cancer cells don't have the metabolic enzyme anymore. So these are interesting because normally under normal circumstances, both cells do fine in terms of normal cells. Still can catalyze the reaction and cancer cells as well because there's still an intact copy of the enzyme. But an interesting thing happens because we can perturb these cells. And then a perturbation of interest to us is the one where we introduce targets and selective drugs to the environment. And then here I'm showing you a hypothetical drug represented by this orange hexagon. And then what this drug does is it basically inhibits enzyme 1, but not enzyme 2. Meaning that if you give this drug to the cell, it will inhibit enzyme 1, but leave the enzyme 2. And then it will do its job. And then these type of drugs are, of course, available and are interesting because they create an opportunity for us to use them in therapy for particular contexts. And the particular context, and that's what we call a vulnerability, we are looking for these scenarios where, as I mentioned, and this is just from my previous slide, I'm showing you normal cells and versus cancer cells. And now you can see that the cancer cells due to homozygous deletion lost one of these enzymes. And now, if you put the drug in context, what you see is that in normal cells, you're inhibiting one of the enzymes, but the normal cells still have the second enzyme, so they can catalyze the reaction. But in cancer cells, an interesting thing happens, and since they lost one of the enzymes already due to deletion, and the other one is being inhibited by this drug. And of course, if this reaction is essential to the cell, what will happen is that when you introduce this drug to the patient, if you use this drug as a therapy option, this drug will selectively kill these cancer cells. And since it's not doing catastrophic effects on the, it's not causing any catastrophic effect on the normal cells, it means that these type of drugs will have really reduced toxicity, so our ideal therapy options. All right, so this is the vulnerability that we are talking about. And as I will mention, we are looking for these type of vulnerabilities in a systematic manner. Our project basically sits in the middle of this pipeline, so this is a computational pipeline that I'm showing. And as you can see, our method stands on the shoulder of really giant databases. Specifically, we are taking genomic profiles from cancer samples, and detailed information about metabolic pathways, and also associations of target drug information, drug targets informations, and now using them as input to our methods. And as an output, we are creating a list of metabolic vulnerabilities that are tailored for each patient, so these are individualized metabolic vulnerabilities associated with drugs. And furthermore, we are also taking CCLE, meaning the cell lines into account, and now we are trying to match each vulnerability with a cell line so that if you're interested in validating the cell line, this vulnerability, you will have the chance to get the cell line and then do an immitial testing. So this is what we do, and this is what I'm going to talk about. And I won't be going into the details too much because processing, and that goes out with the processing and integrating the databases, but the type of resources that we are using are as follows. So first of all, as I mentioned, we are using pathway information for metabolic reactions that are happening inside cells. And for this, we are using this integrated database called Pathway Commons 2, also maintained by our group. We are also including keg enzymes just to increase our coverage and then to be able to capture more metabolic reactions. We are getting all the genomic data, all the cancer genomic data from C-BioPortal. We are using C-BioPortal's web API. So whatever it is in C-BioPortal, we are including it as an input. And as you can imagine, most of these studies are TCGA. We also have CCLE, just to match the vulnerabilities with the cell lines. And we also have these other cancer studies from memory and from other resources. And for the drugs, by the time we started doing this project, there was no resource that was basically combining all of these drug targets information, meaning that which drug targets which protein or which gene. So we had to come up with a tool, and then this is what we call the PyHelper tool, which aggregates data from multiple resources and gives you a list of drug targets relationships. So we established this pipeline. We came up with ways to integrate all of these data and then ran this analysis on 16 cancer studies. One of them is CCLE, as you can see here, and it has the highest number of hits in it, because it is a huge cancer study with 1000 samples in it. And also the data type, of course, is different than the cancer types. But one thing to appreciate from this picture is that, as I mentioned, there are 15 cancer studies and also the CCLE. The number of hits that we find in each cancer study varies based on the number of samples included in that cancer study, or the type of tumor that we are working with, because some of the cancer types are driven by copy number alterations. But these numbers vary, but what you can appreciate that for most of them, we were able to identify vulnerabilities on the order of hundreds. So it is interesting, but looking at this graph is not that interesting, of course. So now I'm going to give you this summary and then give you a sense of what kind of vulnerabilities and how we are doing in terms of the coverage. So on the left, you have this figure, this cartoon. Here you have a hundred figures. If it's a petri dish, then it means that we are including the cell line. So this is all the samples that we are screening in our analysis. Otherwise, it's a tumor sample. So if it is red, it means that we were able to identify at least a single vulnerability in those set of samples. So one thing I didn't mention that in these 16 cancer studies, we have 6,000 samples, 1,000 cell lines and 5,000 tumor samples, so that each figure represents almost 60 samples in this picture. So if it's red, it means that we were able to identify a vulnerability for that set of samples. As you can see, for the cell lines, almost half of them, we were able to identify a vulnerability. And when you look at the cancer samples, you will see that almost 20% of the samples has at least one vulnerability, which is really interesting. And of course, these are two different data sets, so finding a vulnerability in a cell line won't matter that much. But when you look at all the homozygous deletions that lead to a metabolic vulnerability, then an interesting thing happens, because now you see that out of 260 homozygous deletions that cause a vulnerability, a majority of them, that is 150 of them, are shared between tumors and cell lines, meaning that if you were able to find a vulnerability in a cancer sample, it's really likely that you will also find a matching cell line that you can do an in vitro testing for that particular drug. So now I'm going to show you another overview of the data that we have. And now, instead of showing you samples, now I'm showing you sets of vulnerabilities that we identified in this study. So overall, we were able to identify almost 4,000 vulnerabilities across cancer cell lines and also cancer samples. And here, each drug figure, we have 100 of them. Each drug figure represents almost 4D therapeutic vulnerabilities that we identify in this analysis. And then with the color of the drug, I'm showing you what kind of a drug that you can use to target those vulnerabilities. So if a drug is orange, it means that it was FDA approved for cancer therapy. And as you can appreciate, almost 9% of the vulnerabilities that we identified in this study can be targeted by a drug that's already being used in cancer therapy. And if the drug is green, it means that it is FDA approved, but for a different disease. But also, you should appreciate that almost 40% of these vulnerabilities can be targeted by an FDA approved drug, although they're not ideal as cancer drugs. There's still something that means that they have some safety data associated with them. And the rest, colored by gray, means that there are experimental drugs that no FDA approved happened for these drugs. So we created this list, but of course, it's really hard to show what a vulnerability is. So it has multi-dimensional data to it. So we had to came up with a website, supplemental website, that you can use to get overall statistics, see what a vulnerability is. So this is our website. You can get all the vulnerabilities by the frequency, how frequent they are. You can see what kind of genes involved in that vulnerability or the metabolic context of it. You don't have to put down this URL. You can simply search for our title, and then you will land on this page. But you can use to, as I mentioned, you can use this website to get a sense of what is your overall statistics. You can also drill down to a cancer study. For example, here I'm showing four sample vulnerabilities from ovarian cancer. And here we have four patients with four different vulnerabilities. And we are showing extra information for each of these vulnerabilities. And if you're interested in, you can always get more information about a vulnerability, meaning that you can see the pathway context of it or what kind of drugs that you can use to target that vulnerability. Furthermore, we had to came up with a confidence score for each vulnerability, meaning that how confident we are saying that this vulnerability is real in that patient. Of course, I don't have the time to go over all of these details, but it's basically a score that you can use to prioritize the type of vulnerabilities that you want to validate. So we put all of these information as a website, as I mentioned. And then this is published as a computational resource. So we are actively looking for collaborators. But the reason that we did this analysis is that because there is this phenomenon right now that has something to do with the basket trials, of course, we can imagine a cancer patient walking into a clinic where we get the genomic profiling data and feed it into one of these computational methods that are something like I just described. And I get a patient tater that's meaning that individualized list of metabolic vulnerabilities or any type of vulnerabilities for that patient. And in the meantime, if it's possible, you can also establish scenographs or primary cultures from the same tumor material. And now using this list, you can try to imitate or imitate these vulnerabilities. And if you are lucky enough, and if you can show that some of these vulnerabilities are indeed vulnerabilities, you can now go back to the court of patients. And in a basket trial manner, you can collect all of these samples that have the same vulnerability and hopefully establish a clinical trial for them. So finally, again, I would like to thank you for listening to this project. We recently published our results as a computational resource. As I mentioned, we are actively looking for collaborators that we can collaborate to systematically test these vulnerabilities. We have a supplemental website that you can use to explore the vulnerabilities. And our posted number is two, although it's an even number, I'll be around. So feel free to come by and ask any questions if you have. Thanks. We have time for a question. And of course, I was rushing to it. So I forgot to thank Chris, who mentored me during this process, and also Niki Schultz for his really incredible help with the study design. I had a question. That was really fascinating. When you show the number of vulnerabilities, it looked correlated with the number of samples a little bit. Can you give just a sense on what cancers have the most as far as the rate of metabolic vulnerabilities? Right, that's a great question. And I think, yes, the number of vulnerabilities is that we find correlate with the number of samples. But there are also different things, because these type of vulnerabilities happen just by chance. So you expect them to be correlated with the number of samples that you have. The more samples you have, the more vulnerabilities you can identify, because these are happening by chance. But another thing is that some of the, for example, cancer types here, for example, ovarian cancer and also breast cancer, these cancers have subtypes that are driven by copy number alterations. So if you have such a tumor subtype, then you are more likely to find a vulnerability rather than a mutational-driven cancer type. So in your pathway database, some of these metabolic pathways are quite intertwined, and there's backup pathways if you interfere with one or the other. And I assume you haven't gotten to that level complexly. This is just whether there's two enzymes doing one biochemical step and you've lost redundancy there. Exactly true. Yeah, so we are relying on pathway commas, which is curated, manually curated. So we are basically relying on people's idea of what a metabolic reaction is. For each metabolic reaction, it's really easy to grab all of these enzymes that catalyze the same reaction. But of course, there are inconsistencies where you can find two enzymes, although they are labeled as doing the same thing, they can do different things. Or sometimes you have these pathway level alternative pathways that we are not currently considering in our analysis. But yes, that's true. Great, thanks a lot. Thanks. OK, next up is Hui Ding from USC. He's going to be talking about recurrent epistasis defines tumor methylome differences.