 Hi everyone, so I'm going to continue the same from the last speaker to which is a little different from the Other speakers today to talk about rather than the detailed analysis But rather than the application of TCGA data on the drug discovery particularly want to talk about the integration integrating analysis of TCGA data in the identification of normal targets and prediction of patient population for the antibody drug conjugates So I'm talking about the drug conjugates is a new modality that for the treatment of cancer so it consists So it consists of the antibody which is specifically target to the antigens that are sparse express on the south surface as Well as the sat toxic drug which we also refer to as a payload These two components are linked together at by the bar rinker molecule So antibody delivers this a cytotoxic drug into the cancer cells by bounding to the antigens that are specifically Express on the south surface of the cancer tumor cells upon bounding this complex or internalized into the tumor cells through several stages of transport and Indosome vesicles these complex arise at arriving at Lysosome compartment and in this compartment then this Indices will be antibody and also the linker will be Cleaved to release the cytotoxic drugs and this is a cytotoxic drug then exits from the Lysosomes and then reach to its the intended targets either which is DNA or tubal links depends on the Mechanism of action of the payload So currently there are two ADC's on the market Which are targeting CD 30 or her to which most of you probably already heard of Including both the hematopoietic malignancy and the solid tumor In addition, there are more than 30 probably this is a counting by in 2013 by now probably there are more than 30 on the very stage of clinical trials So it's a very active field of research for the cancer treatment So the efficacy the final efficacy of ADC depends on the efficiencies of several steps after the administration of ADC's and Before it reaches its final targets either to bring or DNA or some other Intended targets So the first two steps are the determine the characteristics that Will determine our target selection strategy for the first step after the administration of ADC the Efficiency of percentage ADC's that eventually reaches the tumor cells depends on the bowel distribution of the both the target and also the ADC itself so this requires that this Potential target have very low expression in the normal tissues. So we don't lose ADC's In the normal tissues and also lower the potential toxicity and then the second step is How many percentage of ADC will bound to the tumor cells? This depends this will require us to have the targets that have very high expression on the tumor cell surface So as you can see that in order to increase achieve the optimal therapeutic index for ADC therapeutics We the target should have a characteristic of low expression on the tumor on the normal tissue But high expression on the tumor tissues So of course the protein expression will be the optimum data to you to use for identify such targets however, the port availability of protein expression data is far less than availability of Messenger expression data in different across different tumor types and also different normal tissue type cell types or tissue types so we Resorted to the Messenger like a level expression as Approximation for the protein expression to estimate the target expression on the tumor cell surface as well as the normal tissues So here I'm good I'm just showing you example Although we all know that the correlation of messenger RNA and protein expression level are not perfect But in some cases that they have have shown to be have a very good correlation here is the example between the heart to Messenger expression level and the her to protein level using the TCG data. We see a Very good correlation In addition in the recent clinical trial we have seen that higher response rate to the TDM one Molecule that targets her to show the Absorbed for the patient with a high expression of messenger RNA level by QPTC QPCR. So this is a give us a Some hope that we can potentially use a messenger RNA level expression to as approximation So we decided to use and giving the availability of large scale of TCG data So that we can utilize to estimate the expression in the tumor tumor samples And also the another large scale data set GTACs We can use to use for the normal tissue target expression estimation so in addition to the antigen characteristics another factor that Garnes our the strategies for target identification selection criteria Depending on the ADC payload itself What kind of a mechanism action each of the different payload class have a will govern us to have different selection criteria for example The two very well-known payload class are tubulin based incubators And also the damage based Incubators they have distinct the mechanism action tubulin based incubator mostly targeting the proliferating cells therefore they can Tolerate a lot of medium normal tissue expression. However They're compared to the DNA damage based payload their potency is Less so we require high-level expression of target expression on the tumor in contrast the damage payload class class of payload Targeting both proliferating and non-proliferating cells. So this will require very stringent Expression on the normal tissue expression to lower their toxicity because they are highly potent payload class However, due to their high potency, but we can talk we can have target that have lower level expression so this side just described the strategies that we design to like target for ADC Antibody drug conjugates So from the tumor expression data from TCGA we calculate tumor scores I will describe a little bit more in detail in the next slide But then we also calculate normal score reflecting the target expression on the normal variety of normal tissues using the GTACs data and we formulate different Selection criteria based on the tumor score and normal score these Criteria are tailored to different type of payload class for ADCs In addition because this has to be this type of type of target has to be expressed on the south surface So we also include in prediction based for the transmembrane protein So ensure they are on the south surface So tumor scores are calculated in two there in two factors So the one is that we require these tumor versus normal differential expression to be significant for these potential targets and also using RNA-seq data we can we are able to quantitatively estimate the abundance frequency in other words we can estimate the prevalence of the target expression in certain tumor types using a percentage of samples That are expressed above certain expression threshold. These expression threshold are defined differently for different type of payload class with the Expression threshold is higher than in the macron tubulin based the payload than the DNA damage Similarly, we calculate normal score to reflect their expression in the normal tissues using the GTACs data These are calculated as number of normal tissues express these certain targets at Above certain threshold So this is an example in breast cancer How can we how we can use this two type of scoring systems to identify potential targets? So left panel shows the distribution of normal score versus the tumor scores of all the human genes And then we you upon using the filter of a transmembrane prediction We limited them to these brain dots here and then further by using the tumor score Selection criteria and normal score criteria. We can separately predict the different type of targets for different type of a payload class for example the purple region here We have we have a very high tumor score cut off by the medium normal score cut off This will be targets for the MTI payload which are shown on the right corner here We which we can successfully predict her to which is a known ADC targets In for the micro tubulin based similar type of mechanism based payload so the next example is in the kidney renal Clear cell carcinoma again, this is a distribution between normal score versus a tumor score in this case I'm giving example for the targets that are for DNA payload a damaged payload class so this class requires very low level of normal expression, but Here we can relax on the tumor expression level itself So then the examples are given here. We can predict about a hundred Kind of targets including a known target CD 70, which is you which uses a damage payload base class payload so in addition to identify a very optimum target that are expressed on the south surface And also in addition to develop a very potent ADC another key component for developing a successful ADC therapeutics is to have a right patient. This is a particularly important in for that ADC is targeting the solid tumor because the expression are very Hygrogenous in the solid tumors almost all the in all cases that requires the companion diagnostic For selecting the right patient. So traditionally these biomarkers are based on target expression especially using IHC or ish technologies to estimate the patient the expressions in different patient population Here I'm going to give example to use a TCGA RNA-seq data to predict which patient select a patient population for the ADC targets But also we start to using the mutation data to explore which is a very exploratory Study to see whether we can use the genetic based biomarkers as an alternative biomarkers in addition to the IHC based biomarker Reason is the genetic based biomarker obviously very much more stable In because a lot of the tissues that we use our uncoverable tissues the protein are not as stable as the genetic markers But those are still in the early exploratory Analysis so this is an example how we can use the RNA-seq data from TCGA to estimate the particular target populations for intended ADC targets Using for two as example Similarly to the target expression we used both the differentiation between the normal tissue and the tumor between the tumor tissue and normal tissue as you can see the her tool is significantly expressed in the her to positive tumor subtypes breast subtypes in addition we also estimate is absolute expression level in different tumor types and as you can see again in her to positive patient population Who has a very high expression as we all expected So in addition to the traditional histology subtypes we also want We also thought we should look into some other segmentation of population using the Mutation data that are available. So here is just as an example to look at the EGFR, which is also a known ADC targets Into in different In addition to the known subtypes in breast we also looked at the mutation segments in different Long cancer long I don't know and long scum as well as ovarian cancers to see whether they are overexpressed in certain mutation segments as you can that in agreement with the Literature that EGFR is overexpressed in the long scum as subtypes But in addition to this when we look at the long adenocarcinoma if you look at the overall expression they The mutant the long I don't know tumor doesn't seem to be differentially with the normal tissue But when we look further segment into the mutation EGFR mutation segment, you do see that differentiation between the mutant segment and then the wild type segment So we then look to decided to look at this Genome wide to see whether there are any mutations correlating to a certain ADC targets again using EGFR as example here we also we identify the several mutations that are Significantly correlated with EGFR High high mutation of EGFR most of them are very low frequency Except to the EGFR mutation itself. So this is a scenario here As you can see from this box plot again that EGFR expression is High express high express highly expressed in the mutant mutant EGFR population compared to the wild type population So this gives about 13% In the lung cancer So this data looks promising that we may be for certain for some of the for certain targets Genetic marker maybe can be used as this alternative marker so in summary we have used the TCG RNA seek data to define computation strategy to identify novel ADC targets and also the resource of RNA seek data from TCGA is Axon resources for us to estimate the absolute abundance level of a target because this is a very key factors for The optimal ADC targets. This provides Also guidance for the companion that does development as well as predicting the potential Patient population for the clinical trials. We also explored to use a genetic based biomarker in addition to the protein Expression based biomarkers as alternative biomarkers, which is the more stable The data from the EGFR looks promising So lastly, I would like to send people who has a major contributions to this project This is under leadership of Pareto and yet Vega and my colleagues from the computational biology group as well as my other colleagues from the biological target development target identification and as well as the ADC development group and lastly, I would like to thank TCGA for making such a wonderful resources available and Will and it has and will make great impact in the drug discovery and thank you for your attention To clarify because I might have missed it for the thresholds Was it within tumor types or was you defining your thresholds across the different tumor types for whether? The tumor, you know the the continent you're looking at within the tumor samples within the tumor type So it's in tumor type. So if a tumor type happens to have very high expression of a given gene We didn't know it wouldn't make your cutoff would it because that's because the cutoff is based on all the genes Right, so this is comparing to the median level or the genes. So yes another the single gene. Yeah, so it's Transcriptome level rather than the gene level. Yeah, so just to make sure it's a comparable across the different tumor types So I was just wondering if Clonal evolution would be a problem if you're using a drug an ADC against a particular receptor and Next thing, you know the tumor evolves so that the receptor is no longer prevalent on its surface Has that been observed? Is that a potential problem? Yeah, that's a great question. It has Being our Constant argument right in the ADC field because they may have heterogeneous Expression of certain targets. So one thing we also noticed is Although the the target the ADC is targeting specific targets but we also would look observe the standard standby effect where the actually neighboring cells has been shown to be be able to be killed by the Release the cytotoxic agent. So those are potentially can can be useful for the, you know for those heterogeneous expressed tumor samples But of course the I think the most ideal ADC target are those the heterogeneous I'm sorry homogeneously expressed the targets, you know you're targeting a lot of populations But in addition to that to add on that is also we can we also looking into a combination therapies with other therapeutics, so that will Hopefully If they are expressing the normal cells then they will be going to the normal cells Right, but in terms of different payload right depends on the payload to if the payload is only targeting the proliferating cell For example the micro tubulin based that payload they will have less effects on the normal tissue Then the you know the cancer tissue because I mean just give example We have used GTAC to predict like certain tissue may have Toxicity right, but you know in the you know in the real experiment They may not they may not show those and in some of the targets the known targets right in the mark in on the clinical trials So have high expression in the normal tissue, but they are they're fine in the clinic So they depend it's not only depending on the target expression It also depends on the you know the payload expression payload mechanism itself and also some other Criteria's I didn't touch upon for example how well they are internalized into the tissue So our last speaker for session to this afternoon session is Taya Kleinberg from Institute of Systems Biology Who's gonna tell us about mutation hotspots associated with gene expression signaling pathways protein domains and drug response?