 Welcome to round two this afternoon. We now have our panel discussion titled The Evolution of Cancer Treatment, Immunotherapy at the Forefront. And the moderator for the panel this afternoon is going to be Professor Dorei Swami Ramakrishna. He's the HC Pepper Distinguished Professor in the Davidson School of Chemical Engineering. So they take it over, please. So it's my pleasure to welcome you all for this discussion. One of the appendages to the Distinguished Lecture Series is this exciting panel discussion each time on some burning topic of the day. So today it is on the Evolution of Cancer Treatment Immunotherapy at the Forefront. And if you don't mind, I would like to start off with saying a few words about our distinguished visitor, although there's been, I don't want to embarrass him by more haunting on his behalf. But the issue is that Professor Chakraborty, Professor Rakesh Agarwal and myself have come from the same academic institution. I was a faculty member where a group was a student, but much later after I had left IIT Kanpur. And so the high point of the day was that Rakesh summoned one of his students to take a picture of Arup, Rakesh and myself. The one significant absentee was Rakesh Jain, which is what Arup pointed out. So we'll get with the panel discussion and start off this with a very germane topic of the day, the immunotherapy angle approach to treating cancer. So we'll start with a brief overview of the current role of immunotherapy and introduction. I will do the introduction of panel numbers. This is the panel that you see starting with. We have been imposed on Professor Chakraborty to make any formal statements, unless he feels provoked by the presentations of any of the panelists and he wants to make some observations, he's absolutely welcome to do so. So notice that this is the different approaches that are part of cancer immunotherapy, basically leveraging the immune system to fight cancer. Get that to do it rather than another kind of drug. So this is how the evolution of cancer therapy has happened. It started with chemotherapy, but now it's moving towards the targeted therapy and immunotherapy together. And we will see what the different panelists have to say. We'll begin with Sandro Matusovic, who is an assistant professor in industrial and physical pharmacies, Center for Cancer Research. He'll be addressing the immunotherapy of solid tumors with genetically engineered natural killer cells. This is his area of work. The lab targets immuno-metabolic reprogramming of immune cell function in solid tumors by immuno-engineering cells to redirect inhibition into sustained anti-tumor functions. A widely targeted solid tumor in the lab is glioblastoma, highly immunosuppressive with no treatment. Thank you. My name is Sandro Matusovic. I'm an assistant professor in industrial and physical pharmacy and my lab works on immunotherapy. As you've heard now, there are many types of immunotherapy. All of them target various types of immune systems. So are antibodies that target checkpoint inhibitors, there are vaccines, but also there are cell therapies. And this is the area that my lab works specifically. We work with one particular type of cell. It's called a natural killer cell. This is in contrast to what you might have all heard and be familiar with T cells and CAR T's are therapies that are FDA approved. And so another type of cell is a natural killer cell. Usually, the type of therapy that we particularly are focused on is called a lab-to-cell transfer. This means we take patients, so patients with very severe cancers, we take their blood and we isolate the cells from these patients. We genetically engineer them outside of their bodies and we use various immuno-engineer against ethylbiology approaches to re-engineer these cells to be much more potent at fighting the cancer that we're not able to fight before being while we're still in the body. And then they get re-infused back into the patients. When it comes to natural killer cells, they're present in all of our bodies. We all have them both healthy individuals and sick individuals have a natural killer cells. They're always on, so whenever there's a pathogen, it doesn't have to be cancer. It can be any kind of infection as well. These cells attack specifically for cancer. They attack the cell and then they use various types of proteins on their surface. They're called inhibitory inactivating receptors to kill this cancer cell. The interaction looks very much like what you see in the picture. The little cell is a natural killer cell. The large cell is a tumor cell. They have various different types of antigens. The cancer cells are recognized by the natural killer cells and the NK cells then release something called perforines, which are small proteins that essentially pierce the cancer cell and kill it. This has worked quite well in a lot of blood tumors, so we have been able to treat leukemias completely in people that have been refractive to multiple rounds of chemotherapy. So this therapy has also been approved by the FDA. There are two currently current products on the market that are based on genetically engineered T cells that are called CARC T's and were approved just about a year and a half ago. So we have evidence that this works in patients for whom nothing has worked before. So I mentioned blood cancers. There are also a lot of other tumors that are not blood specific, right? There's brain cancer, breast cancer and a lot of solid tumors and they look more like this. There's no individual cell anymore. Now you're dealing with a tumor mass that usually has to be removed by surgery. It has to then if surgery doesn't work, you have to hit it with chemotherapy. Chemotherapy doesn't work. You have to hit it with radiotherapy sometimes in combination, sometimes at the same time. When it comes to the immune system around the solid tumors, it's very difficult, right? Not only because there's a lot of inhibitory molecules that are floating around the solid tumors but also because getting into a mass like that is really difficult for the cells. They can penetrate. And so even for the tumor that can be targeted with specific targeting cells, getting into deep tumors is very difficult. And this is reflected in clinical data. Any CAR T trials that are currently playing this is about a year old data or so. Everything that you see in gray on the slide is no response. So any type of cancer that we really haven't had any response with immunotherapy yet with cells. And you'll see a lot of them have had nothing in terms of disease that has been able to be controlled. That speaks to the complexity of the tumors. These are very heterogeneous, highly mutative tumors but also speaks to the fact that antigen recognition may not be sufficient to be able to eradicate very difficult tumors such as brain cancer or breast cancer for which surgery and other things haven't worked. And again, a lot of these tumors have been tested after inpatients, after they've already tried chemotherapy, after they've tried traditional radiation and traditional medicines. What do we do in order to re-engineer? So now we're talking about the engineering, immunogeneity of these cells. What do we do? And this is something that we do in our lab. There are gene fragments that we synthetically make in the lab that can redirect cells. So in our bodies, natural cells are going to naturally find a cancer cell and go and kill it. When it comes to complex cancer, they are no longer able to do that. They are very inhibited, cancer is much more potent. So what happens in a very complex tumor, we have to give them a boost and the boost is done by engineering, right? We use genetic engineering to target these cells specifically to recognize specific pathways, antigens, proteins on the surface of cancer cells. And this, like I said, has worked very well in blood cancers, and solid tumors is difficult. One of these types of genes is called a chameleon receptor, but we have also other types of genetic tools that we can use to sort of completely redirect these cells. After they're engineered, we give them back to the patient as a sort of a genetically engineered cell. One specific path we just very quickly in our lab that we target is a lot of solid tumors are characterized by hypoxia, that's low oxygen. Low oxygen triggers the expression of a lot of enzymes and cancers that are inhibitory, and they release metabolites. One of the metabolites is adenosine. Every solid tumor has something what we call adenosine fog, right? It's some type of adenosine that's extracellular released and it just sort of floats there and every time it finds an immune cell that's trying to attack this cancer, it inhibits it. So we wouldn't block that. And one of our current projects in the lab is to re-engineer these cells to overexpress proteins that are usually blocked due to adenosine and then to block CD73, which is the enzyme that causes the production of this very inhibitory, very inhibitory metabolite. And CD73 is expressed a lot of cancers, brain cancer, breast cancer, prostate cancer, and a few others express very high amounts of CD73. So we can engineer these cells genetically in the lab very efficiently and we also have shown that this genetic engineer cells can be re-infused back into mice and this mice, this is a lung cancer model, can very effectively target, can very effectively lead to inhibition of the tumor growth compared to just an antibody treatment alone. Meaning that the boost that the NK cells that have been genetically engineered give this immune system is quite significant. And we also show that these cells can penetrate the tumors much more. So if we do chemistochemical staining, we take the tumor out, we section it and we see how deep the cells can get into the tumor. We see that they get a little deeper if you try to block this adenosine production. Tumor infiltration is a big issue with a lot of solid tumors. So this is a big question for us. And just the last slide is, this is also what we do in the lab and this is sort of some, you can be really creative with the immunoengineering, you can be really creative with synthetic biology and you can create really efficient systems that can only target one tumor and you heard from the previous talk, pathogens in cancer particularly is very smart. They can mutate, they can evolve the expression over time and so you can use synthetic biology to sort of try to outsmart it which is what we try to do in the lab. You target two antigens at the same time, you target three antigens at the same time and things like that. So that's really a combination of sort of bioengineering and immunology which is what we do in the lab. And having said that, I will pass on to the next speaker. So the next person who the panelist is the Xiaoping Bao who is one of our latest additions to the faculty, he's an assistant professor. Basically the interest of the Bao lab is to use innovative technology to engineer human stem cells as models of human development and disease as well to develop cellular and molecular therapies as next generation therapies for degenerative diseases. One focus is the production of all the shelf immune cells such as D in natural killer and I guess you can sort of see this for yourself and I'll let Xiaoping start his presentation. Thanks, Rinki, for the introduction. So I'm a new PI here and my lab works on stem cell based therapy just using like a stem cell to make functional cells. So immune cell is one of my target. So as you know, Sandra mentioned in his slides there are so many like a hot error called cardiac therapy. But if you look at at least cardiac therapy mainly when a patient have like a cancer when he go to the hospital the doctor will kind of take the T cells from this patient and then expand and also engineer the T cell with the cut construct and then expand again and then inject back to the patient. So this is kind of the overall approach to make a cardiac therapy. With that I won't say it's pretty useful especially for leukemia or some broader tumor. But there are still like several major challenges in this field. As I listed here is three. The first one is sometimes it takes like a specific time period to make the cardiac therapy. So it's not good for acute cancer patient. That's one. And the second is cardiac sometimes they have limited proliferation. So when you transplant the T cell may not be enough. And the third one is most of the CAR T now is not so effective in killing the solid tumor. They are really good for like a blood or cancer but not for the solid tumor. So my lab or the field of stem cell field they are trying to at least adjust couple of the challenges. Particularly we are like trying to adjust two challenges. One is trying to make a universal donor cell so we don't have to wait and go to the whole process to make the T therapy. And the second one is with the limited proliferation of CAR T cell. The propellant stem cell has the potential for unlimited proliferation. So in theory you can have this cell universal donor cell that can proliferate without a limitation. And they can be engineered for CAR T and like ready to make the off the show cell product for the patient. So the idea here is it involves two major area. One is how you can engineer the stem cell with the really hard tool, CRISPR-Cas9. You may heard of that. To engineer the stem cell, they are immune response of free for every patient. That's kind of, it's durable and it's just like with a few receptors you need to knock out or knocking a few receptors. That's kind of knowledge from the immune knowledge field. It's durable, I'm not going to cover the detail but with this tool you can just engineer a cell line to make them kind of universal donor. So the second part will involve the direct differentiation. How you can make these cells become the functional immune cells. Like so if you can see maybe not clear, the stem cell go a series of stage to become the T cell. Couple stage is a hemogenic endocelium and a hemoponent stem cell and then go to the lymphocyte progenic and then the immune cells or kind of blood cells as well. So from this kind of human development you can use engineering way to promote the cell differentiation to the functional T cells. So my previous work have established a really robust protocol to make the stem cell to become into the hemogenic endocelium cells. So this is kind of the protocol we've developed in the lab where you just hit the stem cell with one chemical called a CHRR chair, a winter activate. So this is a small molecule without any additional growth factor. You can make hemogenic endocelium cells in five days. So it's pretty rapid and it's pretty cost effective. And the efficiency you can see from the flow sheet on the right, it's positive about 6% for CD31 and 34. These two just a protein marker for endocelial progenic marker and they are functioned to form the tubular structure on the left. And importantly these cells when they culture on the feeder cells called the OP9 which is supposed to promote the immune or T cell and the blood cell differentiation from this population. So as a highlight here on the right figure you may cannot see clearly, but this cell population kind of capable to form the functional T cells and the blood cells. So we are trying to optimize the process and trying to scale up for the CAR T therapy. And that's all what I want to cover today. Chemistry and molecular pharmacology. I think I'm gonna let you see the interest of the LEM laboratory to save time and let him get started with this presentation, yeah. Thank you for introduction. And thank you for having me today because I never imagined the engineering department or colleges has a lot of interest for immunotherapy. So yeah, today I'm gonna talk about my current work. I'm basically working on the breast cancer using the antibody immunotherapy. Yeah, I can skip this cartoon. This one is only show some PD1, PD1 function here. What I found is because everybody knows, in these fields, everybody has some idea about how we develop the antibody for PD1, PDL1, which can enhance anti-tumor immunity because PD1, PDL1 is really well-owned the co-inhibitor receptor. That's why if we block their function you can expect anti-tumor immunity should be increased. But what I found is when I checked the other person's literature or the clinical trials antibody, they didn't pay attention to the protein modification, especially PDL1. When I checked the PDL1 protein, I found they have several post-translational modifications such as post-polation and glycosylation, eukitinations. Among those different types of post-translational modification I found glycosylation or PDL1 protein is really important for their binding affinity. And also the glycosylation itself can determine some antibody affinity or specificity. So we kind of determine the function of glycosylation or PDL1. And then we try to make the antibody which can recognize glycosylated PDL1 only, very specifically. And then we start from the 3,000 hybridoma clones and luckily we have some clone which can block the PD1, PDL1 with very high efficiency. And also some of the antibody can induce the PDL1 protein internalization on their tumor cells. So the internalization means maybe we have some potential to develop the antibody drug conjugate, ADC. So what we did is we add the toxin to the antibody and then treat those PDL1, ADC to the mouse model. And then here, this is the data. Our PDL1, ADC really killed the tumor. Cells in the mouse model. And here is our proposed model. What we did is we just add PDL1, ADC. Of course, those ADC, they can block the PD1, PDL1 interaction, same as other clinical use antibody. And then they can induce the internalization and go to the lysosomal degradation. After that, they can release the toxin. So toxin is, in this case, we use MMAE, which is the tubulin inhibitor. And those toxin can kill those tumor cells directly. Also, we use the cleavable linker between antibody and toxin. That's why those released toxin can go outside and kill other cancer cells, which may not have the PDL1 expression. When you look at the PDL1 expression level in the cell, maybe PDL1 expression is very heterogeneous. That's the reason we choose the bystander effect with the cleavable linkers. And conclusion is, we developed the first immunotropical antibody, ADC, which has a triple impact, which is blockade efficacy and cytotoxicity and bystander effect. Here, at Purdue, I'm working on the PDL1 ADC model. And what we found is, PDL1 resistance tumor still has the PDL1 expression on their surface. That's why we try to make the PDL1 ADC again. And also, what we found is, those resistance cells secrete lots of cytokine. And some of them really contribute to recruit for MDSC, which is the suppressive immune cells. So that's why what we are doing here is we try to combine PDL1 ADC and the blockade identified for to block the MDSC recruitment. This is our lab. And now we are, one person just left. This right now we are five. And we are really enjoying the research here. And then we are making the documentary here. Thank you. So the last of the panelists is Dr. Min Jang, Professor of Department of Statistics, College of Science and Associate Director of Data Science, Purdue University Center for Cancer Research. Nobody can see your specific interest themselves to save time, so please go ahead. Thank you for the introduction. And I'm probably one of the outliers here, as I'm not a biologist. I'm not an immunologist. But I do develop statistical measures that can quite general, can be broadly applied to different type of diseases, including cancer. And I think the one disease we did apply to is arthritis. So I'm going to just share with you some of the methods we have been developing all years. So here is just a diagram that shows that for most of the clinical studies, so we collect the samples from patients, including DNA, RNA, protein, and metabolites. And then we try to look at the data from each of these samples, including the omics data sets, which are getting more and more popular. And the integrism with clinical data, and after the integration, we try to learn a new knowledge from this data and translate it into the practice for general practitioners in the clinical. So the way we are trying to attack this problem is how do we take advantage of this data and try to make that analysis much more efficient and much more accurate. So the one circle is where I'm going to use metabolites as one example and show you how we can do the integration for one type of data. And then I also circle the DNA and RNA to show you the example how we can integrate multiple types of data to try to improve the power of the analysis. So here is an example of the project which is a consortium between IU School of Medicine and Purdue. So we started to look at one of the popular cancer in the state of Indiana as a colorectal cancer. And this started with a collaboration with Joe Peckney. I believe he's from chemical engineering. And Meredith Harrison from one of the faculty researching cancer, you know, research at Purdue. So there are a lot of data collected in this project, including all types of omics, including genomics, metabolomics, glycoproteinomics, and proteomics. In addition to this, there are also demographic profile and clinical data. So our goal here is to try to come up with a statistical model to integrate all this data and which can help us in the early detection of the cancer and monitor the disease progression and to be able to predict the survival of the treatment response. So here's one of the example, like the lower part of the corner with two tables. On the very left, you see there are two metabolizer glycine and serone. So if we look at each one of them and do the statistical test, so the p-value is 0.6 and 0.78, and you know that none of them is even close to the 0.05 we have been using. And so that's the only one that has the smallest p-value is the wildling, which is the last one with p-value 0.01. And then our collaborator was saying that, so this is only the raw p-value. And if you think about there are so many tests you are doing if you control the family-wise error rate, none of them are significant at all. So what we are thinking about is so how we can improve the analysis in terms of the statistical power. So what we did on the right hand side of the table is what if we use knowledge from the biology, like these metabolizers are not independent of each other. So we take the quantity of their correlation between them and our collaborators actually come up with like 15 groups altogether to indicate what metabolizers actually connected with each other. So this is only a subset of the groups. But what I circle here is group 12. And there are only two metabolizers involved, which is the same as the one that circle on the left. And you can see the p-value drops down to 0.005. So this is the power of the integration. So essentially instead of asking the question for each single metabolizer, if we integrate the information and we can actually get much more power statistically, and the clinical did also help us by looking multiple metabolizers instead of one of them because they all contributed to the disease. And on the up top corner, and you see there are like three kind of networks in, so these are the results like follow up with our biological groups. We will try to identify what happens between the health individual, which is a green one and cancer, and also simmering between, we only have the polyps patients, but they are not cancer yet, but they are highly, they have a higher risk to develop cancer later on. And so the connections just indicate whether each of these metabolizers are connected with each other. And then the healthy one, you can see there are seven of them altogether. They are nicely connected. And then when you go to the polyp stage, so there you lost some connections in between, but still there are still seven metabolizers. And then when you really get into a cancer stage, so you lost more connections than you only, and you also lost some metabolizers. So this kind of will show as a system level, so if we look at all these molecules, rather than a single independent molecule, we can gain a lot of information and knowledge in terms of the way that we can understand the disease much better. So the second example I want to show you is more like the statistical method that when we look at the data, so instead of the data that you collected for this specific study, we can actually incorporate external information to try to take advantage of the existing database, literature from PubMed or anywhere, and try to improve our analysis. So on the left-hand side, the top corner is our statistical model, and the y-axis is just the response, the phenotypes in the clinic, and the predictors we want to look at. So if we apply this to the genome-wide association studies, so the x will refer to this whole genome sequence, it can be DNA, it can be RNA, and we know that if we do the sequence, we can get millions of predictors. But on the left-hand side, if we look at these individuals, we can rarely get a million individuals for a study. It's just way too expensive, unless I recently heard that as a VA, so there are many like veterans project, but they do have a lot of money to collect all this data, but in most studies, we're just limited by the number of sample size, so we can collect. And you can see the biological pathway that we try to include. So this can be a DNA level, RNA level, or metabolite level. If we look at the database, if we look at the database or any other literature, there are already information related with a certain phenotype we are looking at, and we can actually incorporate this information into our current analysis through what we call the basing inference, and those are the prior distribution we can formally incorporate into the statistical modeling. And then the left lower corner will be the algorithm. So instead of using the full likelihood function, we actually use the meeting and the mode for the likelihood, which makes the algorithm is very easy to implement, and it runs really fast, and it converts very well. And then the lower part shows the comparison of our method with the popular methods for big data analysis, say lasso and adaptive lasso. And so the left panel is in terms of the prediction error, and the two panels on the right is the false positive and the false negative. So we can actually gain much lower prediction error, so in terms of the different correlation, which indicated by the x-axis, because some of the genetic markers, if you're looking at a single nucleotide polymorphism, if you look at genes, so they may be correlated with each other, either the correlation can be lower or high. So we range the correlation from zero to all the way 2.9 in our simulation studies, and you can see the pink and the blue lines are the methods we develop, so it's kind of consistent in terms of the prediction error, but lasso and adaptive lasso start with a very high prediction error, but it drops when the correlation goes higher. And also the false positive rate, you can see the yellow line on the top is the lasso, which is so very popular in the high-dimensional data analysis, but then the label you're probably couldn't see, but it's 0.8, so which means the false positive rate can be 0.8 if you use lasso for this kind of data. And adaptive lasso is doing so much better, but it's still at the level of 0.3, and so our method basically is only around like between 0 and 0.1. So the lower part will be the false negative rate, so all the methods are doing pretty well, so the scale here is actually 10 to the minus three. So the last example I want to show you is in terms of integration of multiple types of all mixed data, because nowadays a lot of the high-throughput technology will allow us to get, say, the DNA sequence and the RNA sequence, maybe as a multiply profile from the same individual or the same set of cell lines, and how do we integrate multiple types of data from the same individual to try to improve our analysis. So the model here we are using is called a simultaneous equation model framework. So essentially we have this y-th indicator, so it can't be the gene. So if we have, say, 30,000 genes and on the left-hand side of the equation it can't be the first gene, and the right-hand side of the equation can be the other 29,999 genes. So basically you ask the question for each single gene and ask whether all the other genes will regulate this gene or interact with this gene. And by the end of the analysis of these 30,000 models we are going to end up with this network at the bottom here. So this is actually from all the immunogenes and the arrows indicate the direction of the regulation and the different type of the lines indicate the strength of the regulation. So essentially since the algorithm is so fast we actually have the ability to do the analysis using a bootstrap method to repeat it like 10,000 times or 50,000 times. And each time we ask the same question for each of the 30,000 genes. And by the end of this 10,000 bootstrap we collect the frequency for each connection or each regulation. So that's the number that, you know, besides each line. So this will basically tell us the direction of the regulation and how strong the regulation is. Some of them you can see it's bi-directional so basically the regulation is two-way regulation. Some of them is only like one-way regulation and we also have the regulatory effect that can be positive like increase the expression or can be down-regulated. So that will be indicated by different type of arrow. And on the right-hand side of the panel is the simulation we did for different types of network. So the top one is a sparse network and the bottom one is a dense network. So for whatever the network is we evaluate the method is quite robust or stable in terms of the power and the FDR. So that's all I want to say and, oh, sorry. Oh, this is all. Thank you. Yeah, so the last slide is basically all my collaborators and the students who worked on all this project over the past like 14 years helping at Purdue and the bottom is the funding agency that actually make all this happen. Thank you. Let's thank the panelists before we open the floor for questions. If you have questions, please raise them. If you have a particular panelist in mind, name the panelists. If you don't, that's also fine. I have a question. Hi, everyone. Thank you. I'm using it. So thank you, everyone. I, there's an issue in the industry that there's a high rate of failure from bench top to animal models to human studies. And I was wondering if you all had any comments on maybe what researchers can do at an early stage to help increase the success rate of your studies maybe from the bench top level onward to human studies? You might do. So this is correct. And I will comment this from a perspective of immunology and immunotherapy researcher. It is very correct. And picking the right animal model is, of course, the critical first step. But it's also very difficult when you start dealing with immune systems that are very, very vulnerable to impatience. You're a very hard time to predict responses that are going to be seen in patients. And that's part of the reason why I think there are high rates of failure. I mean, there are a few efforts that are being made. And I think that some of them who were mentioned today, I think that a better standardization of the tools that we have available could help get better responses. One of the things is there are a few labs. And I think Dr. Bao is working on off-the-shelf cells. We have a project similar project in the lab. Currently, the way that, for example, cell therapies are being administered to patients are autologous. That means you have to be your own donor. That means that each patient is personalized medication to their patient. Well, that sounds really great because you have your own medication that is really difficult on a manufacturing perspective because you really have to make a drug for each patient, which becomes really difficult. And you can scale it up. You have to scale it out. And that's not always economical. We don't have tools right now where these cells can be taken off the shelf just like you go to CVS and you take a drug for headache. Cell therapies are not there yet. But I think the ultimate goal is where we want to be at a place where we can get immunotherapies more controlled in a way that we can manufacture them better. That's a huge risk. To give you an example, Novartis, who is manufacturing the very first genetically engineered cell, it was just one year on the market, they already produced a few batches of the drug that failed specs and they had to inject into the patients for free. One injection is about half a million dollars. So they had to give that to the patient for free because they just couldn't make enough cells from the patient to give the right dose. This is an example to tell you how it's not just failure in development, but also failure further along development. And I think the more we can maybe control our tools, it's going to maybe help us make better drugs. But that's one thing. And of course, you go back to the discovery thing, that's a whole different conversation. Do we have the right animal model? Do we have the right data to predict responses? I mean, there are a lot of things that, of course, have to be done. So for those of you who are working on CAR-T, from what I've read in literature, there's often significant toxic effects using these drugs, specifically like cytokine buildup and cytokine storms. What new methods or ideas do you guys have to combat those issues? So I can answer this. I work in CAR-NKs. It's not CAR-Ts. CAR-T cells induce something called graft-versus-soil disease and cytokine storm, which is what you mentioned, is when the immunological response is so severe and people have died in the clinical trials and it happens to be an issue. Things for that is to precondition the patient. Other solutions to GVHD and the cytokine storm are to not use T cells. NK cells don't induce GVHD as much as T cells. And that's why our lab is working on NK cells. And other labs are doing it, too, because the MHC and HLA, which is the way that NK cells use a recondition of cancer cells, is not inhibitory. It's different in NK cells. Another solution to completely avoid cytokine storm inside a patient is to use off-the-shelf cells. So anything, for example, connects stem cells and you can derive T cells from them. And those cells are allogeneic. Allogeneic means that you can use one cell for different patients and it will not induce GVHD. The reason for GVHD happens mostly is because it's a mismatch. The donor and the patient are mismatched and that causes the cytokine release. So use different types of cells or use lab engineered standardized cells. That's a big effort in the industry to avoid immune responses. Thank you. Thank you so much for a great talk. As an engineering student, I want to ask what kind of advice would you give for an engineering student that is trying to enter in the medical science field or what kind of unique opportunities that are offered to us engineering students? I think they're turning to me because I have more gray hair. So are you a chemist student? So I think you have a unique ability, which is a following, that you have the ability to think about questions in a relatively integrated and seamless fashion from the molecular scale to the macroscopic. And I think medicine requires that sort of thinking that integrates across scales. Not every kind of training provides you with that ability. For example, chemists are very good at thinking about molecules, perhaps not so much about mesoscopic and macroscopic. These are all statistical arguments. I mean, not every chemist is like that. And similarly, physicists are very good. I mean, the kinds of physicists we are talking about are very good at thinking about mesoscopic and macroscopic scales, not so much about atoms and molecules. And I think you as an engineer have been trained to think across these scales and therefore try to connect phenomena across them, which can be extraordinarily useful. So that's a very general remark. But the other point I might make is I should do what you think is right and what you think is exciting. Now what people around you tell you to do is exciting. And at the risk of taking just two more minutes of these people's time, they've described to you some of the frontiers of cancer immunotherapy today. I want to just briefly, in this context, there are so many young people here, tell you about the history of this because I witnessed some of it. When Jim Allison, who won a Nobel Prize in October, Jim was my colleague at Berkeley, and we were both there, and neither of us are now. Jim was trying to understand the problem that nobody cared about at that time, which is how do T cells shut down after being activated for a while? He was not asking questions about cancer. And his graduate student was Max Crummel. And Max found that CTLA-4, actually, there was a predecessor of Max as well, that CTLA-4 was a molecule on the surface of T cells that got expressed and turned on. When that happened, the T cells shut down. They were exhausted. And they had a ligand, which was B7, on the other side. It didn't bind B7, it didn't have. That's when Jim had the idea. After that very basic study, I mean, at that time in T stone meetings on T cell signaling, Jim was the only person working on this, singular. And nobody cared. And then he discovered that, I mean, he immediately had the idea that I can block this with an antibody. And in cancer, nobody cares whether these T cells are not exhausted because you're willing to deal with the side effects as long as it kills tumor cells. That's an example. He wasn't doing what everybody else was telling him to do. And he wasn't thinking about cancer. Same with CAR-T. The whole CAR-T thing relies on the fact that you do not need the extracellular domain of the T cell receptor. I mean, not N-K cell. What the CAR-T cell part? All you need is a cytoplasmic domain to do the signaling. And then you can put whatever you want on the other side. Art Weiss did that purely to discover, does the outside domain matter? He had no interest in casting me a thermo. He was interested in what the T cell signaling pathway does to get this tremendous sensitivity, things that Art and I still work on together. But so you should also think about doing what you think is the right thing. I mean, and not what the world tells you is the right thing. Well, the world is made up of old people. They don't know. I mean, that could be as adventurous as you, and you should be. Thanks very much to all of you for very good presentations and all of your preparation. I had a question about, with all of the learning that has occurred in scientific laboratories and then been translated to pharmaceutical companies, all that learning in terms of drug development and other things, in regard to cancer immunotherapy, can some of that learning be applied to other diseases that for which the basis is also altered immunity? I'm thinking about rheumatoid arthritis as a chronic disorder and something like sepsis in the acute setting, which is more or less immune dysfunction. So maybe it's the two College of Pharmacy colleagues who are best positioned to answer that. Yeah, that's a good question. Actually, I have a chance to look at the whole text about immunology recently because I have to teach this semester. So that's why this is a great chance to overlook some of the immune system and also immune disease. But I'm not the immunologist. I'm working on the cancer biology a long time. And my major is only belong to just one chapter. We have 17 chapter. And then just last chapter, talk about cancer immunotherapy. But the other one is talk about autoimmune disease or other hypersensitivity issues things. But what I learned from those kind of textbook, now, many big pharmaceutical companies jumped in to develop their batch of antibodies to target the co-inhibitory receptor and the stimulatory receptor. That's why now we have tons of tool to activate or inhibit specific immune cell or specific types of cell. So those things can be applied to many different types of hypersensitivity-related disease or autoimmune disease. So that's why now is great time to apply our finding to other types of immune disease. And I will add that, yes, I agree with all of that. It absolutely can be. And I think that people are looking at that already. Immunotherapy is not just T cells and NK cells. It's also genetic cells, macrophages, MDSCs, a lot of types of cells. So if anything is taught us anything is to really understand that the immune system can be completely re-engineered to be re-functionalized into very different ways. And this is what immunotherapy has taught us over the past 10, 20 years, and especially now. So can it be used absolutely? People are looking at retargeting these cells for various other diseases. I mean, related to the previous seminar, HIV, CAR-T cells are being looked at for HIV treatment as well in a different way than they're done for cancer. But the potential is absolutely there, yes. Thank you for your presentation today. Small problem about the CAR-T cells. So from what I've read and understand from the literature, that the degree of the CAR-T cells has been a kind of rising issue right now. So especially to like solid state tumors where CAR-T cells could not penetrate deep enough into the center of the cells and to totally kill it. So what would you think would be some potential solution for that in the future? The CAR-T therapies that are currently approved on the market have worked for blood cancers because they rely only on antigen recognition. That means the only thing that is is the, it's like a Velcro, right? They recognize that antigen, solid tumors, antigen recognition is no longer sufficient because antigen recognition is not going to be able to allow the cell to penetrate. There are multiple combination approaches that people are starting to look at for solid tumors. To, you know, it's a very extensive topic, but things that people are looking at are co-targeting multiple antigens together with chemokines. These are proteins that allow the cells to penetrate. Also, metabolism is particularly of interest to me because as we look in the lab, there's a lot of metabolic inhibition. Cancer cells use glycolysis to fuel their growth very quickly. That completely changes the landscape around them. So co-targeting that together with antigen recognition is another strategy to help the cells survive a little longer and deeper into the tumor. But yeah, I think it seems to be, at least my opinion, that just targeting antigen is not going to be sufficient to treat, you know, prostate cancer or brain cancer. You're really going to have to start looking at more complex approaches to get the cells in deeper through something alongside, you know, your binding. Thank you for the presentations today and it's very inspiring. And as in nowadays, like we know in our research, the data science is playing a really big role in our research right now. So I want to ask Dr. Jiang, what are the major challenges we're facing in the cancer therapy? And like from all the presentations, I heard like the models where it's tabulation right now is really heavily re-aligned on the high quality and high quantities of the data set. So what if we don't have the resources of those to obtain the high quality or high standard data set? Is there any statistical models for us to utilize to reduce the sample size? Yeah, it's tough. So essentially like, yeah, we do need a relatively like larger sample size to make a lot of a statistical inference. But in the big data world, I think, you know, that's one of the major challenges. How do we deal with data with the sample size is much, much smaller than the number of predictors in your model. Because all the, you know, the statistical textbooks with what we call textbook data is cut nicely with hundreds of, you know, sample size but only kind of a handful of variables. But nowadays with the high throughput technology, we easily generate like millions of, you know, SNP data and thousands of, you know, gene expression profile. And so that's where most of the statisticians who work in the big data world is tackling. So essentially how do we extract the information from this messy and noisy data and get a reliable, you know, knowledge that can keep going for the next step? You know, either do validation in the lab with, you know, cell and animal models or even go beyond this to go to the clinic. So, and there's a big area that we call like, you know, variable selection or data reduction. So there are all sorts of techniques have been, you know, developed to try to, you know, tackle this issue in terms of, you know, either using some statistical methods to deal with big data but you add a penalty term and this penalty term is penalized for the high dimension for the noises in the data and also one of, you know, the AI, you know, algorithms you learn from lots of data from the literature, from the previous studies and you try to come up with a small subset that you think is a reliable predictor that you can keep going. So there, and also the basic inference I was talking about is getting more popular because nowadays, you know, the basic inference is kind of, you know, delayed because we don't have enough computational power and so with the HPC is a high performance computing so this kind of come back to the stage like people, we don't need to worry about, you know, we don't have enough memory, we don't have enough space because we can, you know, use the current, you know, computing technology environment to be able to do a lot of things simultaneously through the parallel computing or through the GPU all these things. So I guess, you know, the area is moving forward with the data being messier and the larger but with the computational tools and some newly developed statistical, you know, methods and the corresponding algorithm is getting there. So I have a question directed to Professor Chakraborty. So in the stochastic model that you showed how parameterized are those models to work with the experimental data and what is the role of the ebony shoe parameters and how important are they for going towards maybe generalizing all the models that you showed? The answer is different for each of the classes of things I showed. So for the inference that we did of the mechanistic of the fitness landscape or rather the prevalence landscape, that of course has no parameters. That is, we were inferring the parameters from the sequences. So that was just a learning problem. And so there are all the points that were just made apply the unless you have enough statistical power, you're not gonna infer the right parameters. But then your question becomes relevant for the next stage that we did, which I'm just doing it by illustration that the mechanistic models that we constructed then to try and deconvolute the prevalence from fitness by thinking about the evolutionary dynamics of population wide immune responses. Certainly there were parameters in particular. The parameter that went into that was the mutation rate of the virus, which is known. And in a course-grain model like that, the other parameter is the strength of the immune response that each person put. And that we drew from a statistical distribution. And then the question is where do different people attack this? And then also we were informed by clinical data on the distribution or probability distribution of where people attack things. They were purely stochastic estimates. And then we found that if we moved those around, I mean changed those around within some scale, then we got similar qualitative results. And the only thing we wanted to test was the qualitative prediction of the order of fitness. So that was one case. And the same holds when we were predicting evolution in a person, it's exactly the same questions. When we do the affinity maturation calculation in the second part, it's certainly parameters. And some of them are known and others are not known. And again, the predictions we were making were all qualitative. If you give a cocktail, it does this. If you do sequence, it does this. And you might say even that could be unrebust to these unknown parameters. And I was, Ramki was asking me the same questions both this morning and right now. First of all, you can never be certain that you are right because you can only think about these questions up to a certain degree of, but what you can see is if the qualitative results seem very mechanistically reasonable, you feel the underlying mechanism is something you feel a little bit more confident. The other is very important is biological systems have what Jim Setna has called sloppy modes. That is that most of the dynamics is determined by modes that have a lot of, not huge, but are relatively parameter insensitive within some range. And there are few which are extraordinarily important. And the reason is that if a biological system was so sensitive to every parameter in this complex network, I mean, she showed you some networks there that were what are called the usual hairball is the analogy. You know, the expression level of proteins in organisms varies quite a bit, not just from one organism to the other, but from you for minute to minute, right? So enormous parameter sensitivity to all those parameters would make this a very unrobust system. And so many of these modes, you can do toy models to actually illustrate that that you will get in dynamical sense, eigenvalues with eigenvectors that are extremely soft. And so there are parameter ranges in which it will still function qualitatively give you the same answer. But you can never be sure, as I also mentioned during my talk, until you test the veracity of what you have done against in vitro and clinical data because, but you need some level of confidence built up for mechanistic understanding and so on and so on. Before you can tell somebody to spend a million dollars to do that experiment. You know, especially monkey experiments, you know, 10 monkeys is a million dollars experiment. We are getting to a point to when we should close this out because it's getting to be the time allotted by 15. I want to thank the panelists again with a little tone of apology because we put this all together in a very short amount of time. We didn't give you the kind of time that probably you needed to prepare, but you still came through and made it possible to discuss this at a pretty good level. Saying that, I need to also add a special note of thanks to Dr. Bill Clark. Who did most of the work of organizing this? I'm just the moderator, just kind of taking advantage of what he did. I want to thank you again. And let's. Thank you.