- This event has passed.
Endless Summer School: Healthcare Roundup
March 17, 2022 @ 10:00 am - 12:00 pm
This session features Vector Faculty and industry leaders who are breaking ground in healthcare applications of AI. Bo Wang will showcase the AI-enabled COVID genotyping tool developed with DNAStack and the Cloud Supercluster Project. This event also features Elham Dolatabadi’s tick identification model for the prevention of Lyme disease, and and much more.
Shems Saleh, facilitator
How machine learning can improve the genomic surveillance of SARS-COV-2 by Bo Wang, University of Toronto, Vector Institute
The ongoing COVID-19 pandemic is the greatest health-care challenge of this generation. Genomic sequencing of the SARS-COV2 has provided a powerful tool to understand the virus pathogenesis with immediate implications for vaccine design, drug development, and efforts to control the pandemic. To facilitate the epidemiological tracking of SARS-CoV-2, an unprecedented effort to make COVID-19-related data accessible in near real-time has resulted in millions of publicly available genome sequences of SARS-CoV-2 on Global Initiative on Sharing All Influenza Data (GISAID). In this talk, I will talk about recent efforts in utilizing machine learning to improve the timely genomic surveillance of SARS-COV-2 with interactive visualization and effective detection of new variants.
Privacy Enhancing Technologies for Gigapixel Medical Image Analysis by Jesse Cresswell, Layer6
Machine learning’s progress in medical image analysis is hindered compared to other fields by a lack of publicly available multi-centric and diverse datasets. The inability to collate medical images from many sources mainly stems from confidentiality and privacy concerns around sharing medical data. Collaboration between healthcare institutions may be enabled with appropriate privacy enhancing technologies, including federated learning and differential privacy. In this talk I will demonstrate methods for privacy-preserving machine learning at scale on gigapixel histopathology images, the largest and most complex type of medical image. These ideas can be used amongst participating institutions to privately share information, thereby increasing the effective dataset size available to each, and in turn improving model performance.
Deep Learning meets Public Health Surveillance by Elham Dolatabadi, Vector Institute
A priority for public health is to enhance the systematic collection, analysis, and interpretation of data essential to the planning, implementation, and evaluation of public health practice. Production of data and the ability to convert those data into usable information at a large scale and in a timely manner can initiate appropriate public health action. In this talk, I will be presenting two recent projects that were conducted in collaboration with public and private sectors on modernizing public health surveillance and response systems in order to harness the power and promise of user-generated data. I will demonstrate our recent effort on using emerging methods and technologies in data science and deep learning to enable timely and large-scale management and analysis of users’ data for two different use cases related to infectious disease.
Counterfactuals for explainability in drug discovery by Ali Madani, Cyclica
Deep learning methods have shown much promise in accelerating the development of novel medicines. While it is possible to use deep learning to guide the drug discovery process, the reasoning behind many of the resulting predictions are complex and often inaccessible to human users. The ability to explain a given prediction in human terms is critical, for both the use of the model, and for its development and validation. Considering drug-target interaction predictions, as one of the fundamental tasks in early stages of small molecule based drug discovery, users often wish to understand why a model might prefer one molecule over another. Developers need to understand in detail how the changes they make can impact individual predictions. To address some of these concerns, we have developed a method for producing molecular counterfactuals. Using the open source python package Deriver, developed at Cyclica Inc, a given molecule can be used to derive a chemically reasonable local space, populated by molecules which are highly similar. Differences in score for predicted drug-target interactions, using technologies like MatchMaker deep learning engine can be assessed, and the results can inform which parts of the parent molecule may be contributing positively or negatively. Understanding the contribution of molecular features via such counterfactual analysis could help the drug discovery community to effectively use deep learning models for small molecule drug discovery.
This event is open to Vector Sponsors, Vector Researchers, and invited health partners only. Any registration that is found not to be a Vector Sponsor, Vector Researcher or invited health partner will be asked to provide verification and, if unable to do so, will not be able to attend the event. Please contact events at vectorinstitute.ai with any questions.
About the speakers
Bo Wang is tenure-track assistant professor of Department of Laboratory Medicine and Pathobiology and Department of Computer Science at University of Toronto. He holds a CIFAR AI Chair at the Vector Institute. He also leads the AI team for the Peter Munk Cardiac Centre at the University Health Network. Bo Wang obtained his PhD from the Department of Computer Science at Stanford University, and has extensive industrial research experience at many leading companies such as Illumina and Genentech. His PhD work covers statistical methods for solving problems in computational biology with an emphasis on integrative cancer analysis and single-cell analysis. Bo Wang’s long-term research goals aim to develop integrative and interpretable machine learning algorithms that can help clinicians with predictive models and decision support to tailor patients’ care to their unique clinical and genomic traits.
Jesse Cresswell is a Sr. Machine Learning Scientist at Layer 6 AI within TD, and is the Team Lead for Risk. His applied work centers on building machine learning models in high risk and highly regulated domains. For research, Jesse leads the Privacy group looking at applications of federated learning and differential privacy to banking and healthcare. He also pursues fundamental research on deep generative models for understanding the manifold structure of real-world data. Before joining Layer 6, Jesse completed his PhD at the Department of Physics, University of Toronto. His research at the intersection of quantum computing and string theory was supported by a Vanier Scholarship.
Elham Dolatabadi is currently an applied Machine Learning scientist and part of the AI engineering team at Vector Institute. She is also an Assistant Professor (status only) at the Institute of Health Policy Management. Elham’s work portfolio and research agenda are mainly focused on the adoption of Machine Learning (ML) and Deep Learning technologies into real-world needs. Elham’s mission is to bring the power of ML and Data Science to health in order to address the big challenges facing our healthcare system. She has over 10 years of experience in leading various health AI projects in collaboration with public and private health sectors. Elham has well-established experiences in developing and teaching health AI training programs and graduate-level courses on ML and knowledge representation in the health domain. Prior to her role at Vector Institute, she was a postdoctoral researcher at University Health Network where her research focused on applying ML techniques in the development and evaluation of intelligent health and safety monitoring technologies. Elham completed her Ph.D. at the University of Toronto focused on intelligent ambient sensing technologies and pattern recognition.
Ali Madani is Director of Machine Learning at Cyclica Inc, a Toronto based biotechnology company and leads a team of scientists and engineers to further improve Cyclica’s deep learning and machine learning technologies for small molecule drug discovery. As a machine learning specialist, Ali has worked on a series of scientific articles in high impact scientific journals and international conferences covering such fields as transfer learning, graph neural networks and representation learning with the focus on healthcare applications. He is also Editor of special Topic Artificial Intelligence In Cancer Diagnosis and Therapy at MDPI. Ali obtained his PhD at University of Toronto focusing on application of machine learning in personalized cancer medicine and biomarker discovery, after finishing his two masters in Mathematics and Engineering at University of Waterloo.
Shems Saleh is currently undertaking product development at a MLxHealthcare startup that’s currently in stealth mode. She’s especially interested in the responsible deployment and implementation of Machine Learning. Shems likes to explore questions related to social implications of Machine Learning algorithms on different people. Prior to this, she was a member of Vector Institute’s engineering and industry innovation teams which work with the institute’s researchers and focus on Applied Machine Learning in health, industry, and government. She received her Master’s and BSc from the University of Toronto. Her Master’s work, done in affiliation with the Hospital for Sick Children, focused on modeling temporal health data. Shems has also been involved in organizing multiple workshops/academic events including FairML in Healthcare and ML for Mental Health.