What: Predictors for Hypermutating Brain Cancer Identified with Artificial Intelligence
When: October 3rd, 2023. 9:30 AM ET
Who: Prof. Jiguang Wang, Padma Harilela Associate Professor of Life Science, Director of Wang Digital Health Laboratory
Where: Newswise Live Zoom Room (address will be included in follow-up email)
Details:
Researchers from the Hong Kong University of Science and Technology have developed machine learning models trained to analyze characteristics of cancerous brain tumors and have identified critical markers to predict the aggressiveness of certain glioma subtypes. Irregularities in these genetic markers correspond to a greater probability of somatic hypermutation, occurring when errors in cellular division cause a mismatch in DNA base pairs. These early predictors of hypermutation, including genes related to DNA replication and transcription throughout the cell cycle, offer a valuable tool for predicting tumor behavior and tailoring treatment strategies.
TRANSCRIPT
Thom (Newswise): Hello and welcome to this virtual live press briefing. Today we have research about a machine-learning model to forecast glioma treatment and the effects. We have with us Professor Jiguang Wang. He's the Padma Harilela Associate Professor of Life Science, and Director of the Wang Digital Health Laboratory at the Hong Kong University of Science and Technology. Professor Wang please go ahead and share your presentation and tell us about your research.
Dr. Jiguang Wang (Hong Kong University of Science and Technology): All right, thank you Tom, for the introduction. So I will start screen sharing. Can you see my screen?
Thom: Yes, we can.
Dr. Wang: Do you see the full screen?
Dr. Wang: Okay, so all right. So let me start briefly to introduce my work on the study of cancer evolution. So the purpose of the work is basically to address neuro-oncology problems. So let me start with a brief introduction of the background. So, as we all know, adult diffuse glioma is challenging human health. So it's one of the most aggressive tumor for humans, especially for adults. So the standard treatment now for glioblastoma, or for short, GBM is named Stupp therapy. But patients usually suffer from relapse. And the treatment for recurrent GBM is still controversial. A recent study actually report that the gliomas have interesting observation in terms of their ratio specificities. We do observe, like younger age in Asian gliomas, and also we observed like a lower incidence rate in their Asian gliomas to study cancer, so including remarks. So actually, recent programs have been devoted to the study of machine learning for large pharma call genomic data, trying to predict their treatment, and using drug repurposing or new compounds to more efficiently treat cancer patients. So while of their recent studies from our collaborators, and our group, is they're using patient derived organoids to, like, generate big data, then we can use the big data to train machine-learning models to study what's the right compound for the right patient. So the recent study has demonstrated the power of machine-learning. So I'm using their big data to like predict their optimal treatment for particular patients. However, this strategy, although very promising, but is having a big problem. We all know the tumor, which was used to study the genomics, to study the drug resistance, it's actually the tumor cells taken from the surgery. So the tumor cells taken from the surgery had been investigated for the purpose of identifying right drugs. However, the recurrent tumor usually is caused by the residue tumors or is driven by the tumor after, like, recurrence. So Okay, question five will be whether the recurrent tumor is different or the same with their primary tumor. If they are the same, then we can record using the strategy I mentioned above. However, if the tumors are different from their primary tumor at a recurrent time, so then the strategy will fail, so therefore, the unit patter, I say, five to six years. So my group, and also some collaborators are trying to address the problem, trying to investigate the process of cancer evolution, trying to understand whether the recurrent glioblastoma is different, or the same with primary glioblastoma. So our recent study showed that so far the recurrent form of tumors, so actually, they are dramatically different in terms of their mutations in like MSH6 and hypermutation. Also mutation in the MYC pathway, mutations in the MGMT and so on. So, those results actually are telling us that the recurrent tumor are different from the primary tumor. Therefore, it's really important to be able to predict cancer evolution from the information we have collected from initial status, so we're asking a question whether we can use the information collected at an early stage to predict the behaviors of cancer cells at late stage. So, by searching literature, actually, there the study published in 2019 by GLASS Corsortium and led by their professor Verhaak, and they found that, from the study of longitudinal glioma from about 200 patients, so, they claim that the evolution process for glioma actually is largely stochastic. So however, we believe, although the process is stochastic, there should be some like factors that can predetermine the evolution path of glioma patients. And our question will be how to use those pressures, longitudinal patients dataset to predict cancer behavior based on a history with others. So this is actually the cartoon to show during the evolution process and their treatment for cancer patient or cancer cells should go to different directions. It could develop hypermutation, or it could develop mutations in MYC or MGMT genes. So how the cells can be predicted this way, we want to use the big data strategy to collect all the longitudinal data in the literature and also collecting a lot from our collaborators in a hospital to use those data, identify patterns of cancer evolution and using these patterns to develop machine learning algorithms to predict the direction of cancer evolution. To do that, so our group and our collaborators together, have integrated data sets from more than 500 patients. For each patient, we have at least two time points. The first time point is named at initial tumor. The second point, 10 points is actually after the treatment from Temozolomide and the standard treatment of gliomas, and we collect tumor again, at a recurrence, then we do the sequencing to understand the genomic change of the cancer cells, before and after treatment. We observe several interesting thing. So one of the reason because of their data set is a little bit different from the public literature. If you check the literature, actually most of the data are from a GLASS Consortium. So those patients are almost 90% from the Caucasian cohort. So collecting from the local hospital in East Asia, so we got over a 100 patients, so the East Asian population, so those patients are going to have a two time points for each patient, and we do the sequencing to understand their genomics. So by comparing, the genomic from East Asian glioma patients to the patient from Caucasians, so we find several interesting thing, but the most interesting thing we have observed is their amplification or the chromosome gain of MYC. So as shown in this plot on the left side, the MYC gain actually have more like founding alteration or Truncal alterations, in Asian areas compared to other Caucasian areas. On the opposite, so if you check the literature, you can find that actually, in a Caucasian cohort, there is a one snip, which is activity in the MYC pathway at chromosome 8q24, and this snip actually is almost zero, almost not observed in the Caucasian cohort, so it's more popular in there. Sorry, it's not observing in the East Asian cohort, it’s more popular in the Caucasian cohort, indicating that to activate their mid pathway, Caucasian patients and East Asian patients have alternative way of activity in the same pathway. So therefore, we believe that this new cohorts could bring us the power to identify a more powerful model. So based on this idea, so we develop CELLO2. It's a model to predict cancer evolution based on Longitudinal data sets. So CELLO2 can actually successfully predict the occurrence of hypermutation, based on the feature, correct, that are 10 points. If we check the importance of the feature. So the number one important feature is the treatment, as a latent agency including Temozolomide. Number two is actually the genomic feature is a MYC thing I have just mentioned. So if we use our model to compare with the WHO 2021 grading of gliomas, we will find out that majority of the tumors actually are classified almost the same with T22-1, which is the most recent classification on grading is pretty much, but we observe some difference in terms of these 30 patients, so they are graded as grade two and three by WHO, but they are graded at the high risk by our maturity model. So, those patients actually have much worse post progression free survival, and those patients carry out recently MYC. So, further indicate MYC amplification could be very important for patient diagnosis and for patient greeting. So that we perform the experimental validation to show why the MYC is important for determining the occurrence of somatic hypermutation. So, we did experiments in patient derived fairlife, which is from Beijing China hospital. So, by this premise, we successfully proved the relationship between MYC amplification and the occurrence of somatic hypermutation after tells on our treatment. So, in terms of magnetism, we tried to overexpress MYC in cell lines and trying to investigate epigenomic change of their cells after the MYC amplification. We found out that the MYC actually open the chromatin, and they increase the probability of their chromatin to be damaged by their Temozolomide. So that increases the probability of their mutation occurrence, and the gene named MSH6 eventually leading to the occurrence of somatic hypermutation in those patients who are taking Temozolomide. So this is actually the work I wanted to share with you. And lastly, I would like to share with you about the ongoing work in the light. So we are computational live, but we are also working on systems biology, trying to understand the process of cancer illusion, and trying to use the process of cancer illusion to help to improve their treatment for cancer patients, especially for brain tumors is about our major focus. Last, I would like to thank our students working on these projects, and also our collaborators from Beijing Neurological Institute, Beijing Hospital, from Samsung Medical Center, the University of Miami, and of course from CUHK, and also our colleagues at HKUST. Okay, thank you for your attention. Any questions will be welcome.
Thom: Thank you, Dr. Wang. Tell us if you can summarize the significance of your research in terms of how and when this can begin helping patients and clinicians.
Dr. Wang: All right, thank you. So let me go back to this site. Basically, this is the highlight we want to say. So for patients and clinicians, so they can right now start to use CELLO2, to evaluate the patient tumor samples if they have their genomic sequencing. So they can immediately input their data into data set. And if left, this will tell the probability of this patient to develop hypermutation and develop progression after the running of the software. So for patients and clinicians, we can use the software to recreate their tumor grading, and firstly we can use this one to have a prognostic prediction. Secondly, approach can better tell the patient whether the patient is suitable for using the current health treatment, or the patient should pursue more advanced immunotherapy or other therapies.
Jessica Johnson (Newswise): Professor Wang is this available to clinicians now at this link? They can access it? They can access the model and put in the data?
Dr. Wang: Right, so it's totally free.
Jessica Johnson: Okay
Thom: We have a question from Forrest Ray. And his question is... is CELLO2 already clinically deployed? Or does it need to undergo further testing to become available?
Dr. Wang: So right now, it's available, but it's not clinically available. I think it's more like for research purpose. And for patients, they can see the results, but like how should they use it directly on clinical purpose, they need to discuss with their doctors.
Thom: Talk about how important it is for recurrence free survival in brain cancer patients and how this model could improve that.
Dr. Wang: Right. So, so far, according to the data we have collected, so the model cannot predict, like a recurrence free survival. So the prediction is based on the poster progression survival, so we can tell the patients, after progression, what is the expected survival time.
Thom: And what's the quality of life and typical prognosis for glioma patients in particular when hypermutation occurs?
Dr. Wang: So, according to our current data, we don't see any significant difference in terms of their life quality. But again, so the difference is under post progression, like survival and maybe also quality, but we do need more data to answer this question.
Thom: In what ways could it help improve quality of life as people explore these sorts of precision medicine treatments, rather than just going with the standard which is now correlated to the possibility of hypermutation?
Dr. Wang: Right. So this is a very good question. So, actually, our model to suggest for the patients who have the MYC amplification, if they are taking Temozolomide, they will have much higher chance to have hyper hypermutation happen. So for hyper mutated cases, I've seen according to their most recent knowledge, so in the immunotherapy, those patients might have larger chance to respond to immune therapy. But we do need to perform a further study to verify this hypothesis.
Thom: So the MYC amplification "MYC". What can patients do? Or how do you recommend patients speak to their doctors in order to get the kind of testing to pursue identify those precision medicine targets?
Dr. Wang: Right, so the copy number gain in MYC actually is rapidly detected by a whole exome sequencing or whole genome sequencing. So if any sequencing was performed, it's also possible to identify the amplification, but the accuracy will be lower.
Thom: Is your CELLO2 model currently being used in any other human oncology studies?
Dr. Wang: Not yet.
Thom: Not yet, but you plan to?
Dr. Wang: Yes. So this is our current plan.
Thom: Great, great, that question was another from Forrest. If there's any other question from our audience, please feel free to chat those to me. We also want to know what you hope to accomplish with working in this precision medicine area overall. Are you particularly focused on glioma? Or are you looking at other areas of oncology? And what's your overall goal with this?
Dr. Wang: So my lab is mostly focusing on gliomas. We’re also, like working on several other intracranial disorders, not only for gliomas, but also for others less aggressive tumors in the brain. Our focus is the brain.
Thom: Another question from Forrest. Will these future studies that you hope to do with the CELLO2 model, will they be in collaboration with any pharmaceutical companies? Or would they be likely entirely academic?
Dr. Wang: Yeah, this is also our plan to work with more like pharmaceutical companies trying to optimize the application of drugs. So not only considering the current status of the mutations, but also the future mutations. So if we can target the future mutations, this could be a way to avoid recurrence. In principle.
Thom: You mentioned the differences between the East Asian and Caucasian brain tumors and especially the genetic characteristics for things like MYC, can you elaborate on those differences in what the potential clinical implications are? Does that mean that if, for example, Asians are diagnosed with glioma, they may be at higher risk to develop the hyper mutating form?
Dr. Wang: Statistically, it’s not different. So it’s just because glioma is too heterogeneous, so it's very difficult to really predict their hypermutation [unclear] based on one mutation. So, however, we do see a lot of difference. So between the Caucasian and East Asians in glioma patients. So the number one difference is actually the amplification in criminal filing and the deletion in chromosome 10. Though, it is more popular in Caucasians, but less popular in East Asian. So another different is actually the one I mentioned, so if there is a snip, so in MYC, so it's actually an alternative row of activity and mix pathway. So in Caucasians, actually there're patients, many patients carrying their mix snip. So but in East Asian "0" patients carried this snip, but those patients have their amplification in the chromosomes. In terms of indications, I do think, for the treatment of gliomas we have now, it's using very standard treatment for all patients. But we do need to consider the potential ethnic difference, before treatment. So maybe in the future, I could be for the treatment, we can test the snip, so the patients, and then we can design better treatment strategies based on the snip and the based on the schema color and based on many other factors.
Thom: If there are any other questions from our audience, please chat them or feel free to turn your mic on and speak up. Dr. Wang, what do you think is the most important thing about your study and this machine learning model for science writers especially to understand so they can communicate it to the public?
Dr. Wang: Right, I think the most important thing here is that our study is actually providing a rich resource for the public and all the scientists,or students, or people who have the interest in understanding cancer evolution, they can freely download the data to play with it. So this is the first thing. The second thing I want to share is for the machine learning models, actually, we cannot think... It's not a miracle for today. So, actually, they are trained based on big data. So the more data we are collecting, so the more accuracy the model will be achieving. So therefore, so we are keep collecting more longitudinal data from more hospitals and more patients, trying to improve the accuracy of the model, and also trying to use the model to help patients to their better management.
Thom: Great, thank you very much, Dr. Wang. I believe that's all the questions we have for today. So I would like to say thank you, and with Hong Kong University of Science and Technology, on behalf of Professor JEE Kwang Lang and his paper coming out in the journal Science later on today. We will have a recording and a transcript of this presentation available for any media who would like to view it on demand and share the contact information so that you can follow up with any further questions. Thank you again, Dr. Wang. Best of luck with your future research.
Dr. Wang: Thank you so much.
Thom: Thank you all very much. Have a great rest of your day.
Dr. Wang: Have a good day.
RSS