In light of the challenges that COVID-19 presents to our society, Mila is bringing its machine learning expertise to the scientific community together with its partners across different disciplines to help find potential solutions.
To join the effort, you are encouraged to reach out.
Contact: Yoshua Bengio (Mila/UdeM)
Description: The objective is to make confinement more efficient by being selective, i.e., allowing the people at lower risk of being contagious to be under fewer constraints while putting stronger isolation pressure on those at high risk. The project is going at full speed with two dozen people involved to develop a tracing app for Canada and Quebec, with a Machine Learning component to predict a person’s probability of having the disease based on contact history and medical information. The project has a strong focus on privacy (aiming at peer-to-peer communication to avoid a central repository of individuals’ movements) and relies on worldwide open-source efforts going on to develop the required building blocks.
Contact: Fred Hamprecht (Heidelberg University), Nasim Rahaman (Mila/Tuebingen University), Florian Jug (MPI), Oliver Stegle (EMBL)
Description: A space for enabling and accelerating collaborations between medical professionals and data scientists.
Description: This family of interconnected projects focuses on medical data collection (omics – genomic, proteomic, transcriptomic, and clinical data) combined with statistical machine learning analysis in order to achieve a better mechanistic understanding of the COVID-19 disease progression. The aim is to facilitate new drug development and/or repurpose existing drugs via the identification of binding targets for antiviral agents and potential vaccines as well as for more personalized patient treatment due to a more accurate assessment of disease progression and risk level. We are interested in viral evolution and viral-host interactions of SARS-CoV-2 strains that may affect humans differently depending on their ancestry, sex, age, co-morbidities, current medication and personal genetic profile. Such interactions may be key for understanding how the epidemic is unravelling and to eventually develop personalized treatments.
Description: Since the beginning of the SARS-CoV-2 outbreak, the virus causing COVID-19 is and has been evolving from its original RNA sequence from a bat viral strain. Using deep learning models and genomic sequences of the SARS-CoV-2 virus from patients, we will aim to detect interaction effects within the viral genomic sequences. In particular, we will apply artificial neural networks to time-sampled whole-genome datasets for feature learning to analyze allelic interactions. Second, neural network approaches can be trained from the large amount of sequences available from diverse geographic regions to predict the possible point mutations that appear on RNA sequences based on the other nucleotides in the sequence. These approaches may reveal mid-range interactions across the viral genome and mutations that arise on specific proteins in connection to treatment response and disease severity.
Project 2: Dynamics of the immune response in young COVID-19 patients
Description: Why do children and young adults rarely present severe symptoms to SARS-CoV-2 infection? What causes some young patients to have severe symptoms? CHU Sainte-Justine, the largest mother and child centre in Canada, will collect clinical samples and data on young patients who are admitted to the ICU for COVID-19 to uncover the molecular and clinical features of these rare immune responses. We will generate, aggregate and interrogate multi-omic data, including patient genomes, virus genomes and single molecule blood transcriptomes with comprehensive clinical data to identify common features of severe pediatric COVID-19 cases. These features will then guide the design of functional studies to dissect the dynamics of the immune response to SARS-CoV-2 infection, with the potential to discover prognostic biomarkers.
Description: The teams of Yoshua Bengio (UdeM), Jian Tang (HEC Montréal) and graduate student Maksym Korablyov (UdeM) have developed a deep reinforcement learning system, LambdaZero, which can quickly evaluate billions of candidate molecules. The approach can gradually modify the molecular structure by adding or removing building blocks in order to converge toward new molecular structures that can bind a target protein. The Mila team is looking for collaborators with expertise in deep reinforcement learning and drug discovery to help with their research.
Description: Using machine learning research at Mila and working with researchers in bioinformatics, immunology and vaccine design, InVivo AI is building an open-source platform leveraging the latest AI technologies to model pathways in the immune system in order to better predict immune response. The team is working on modelling proteasomal cleavage, TAP transport, antigen binding to MHCs, antigen processing to the cell surface and recognition of MHC-antigen complexes by T-cells and B-cells with the goal of providing insight into immunogenic regions of the virus.
More specifically, we work on AI approaches that can contribute to the process of vaccine design in the following ways:
The team is working on modelling the complete antigen processing pathway: proteasomal cleavage, TAP transport, antigen binding to MHCs, antigen processing to the cell surface and recognition of MHC-antigen complexes by T-cells and B-cells. The goal is to provide insight into immunogenic regions of the viruses. We are seeking collaborators across all disciplines to work toward designing better vaccines. Get in touch with us on the #ai-for-vaccines channel of our slack or send a note to Sébastien Giguère (email@example.com).
Contact: Sasha Luccioni (Mila/UdeM)
Description: Sasha Luccioni (post-doctoral researcher at Mila) and her team are working with researchers from UN Global Pulse to map the landscape of current and proposed research papers that use AI to counter the COVID-19 pandemic. The article, published on March 25, covers the many facets of the crisis, including molecular modelling, epidemiology and diagnosis in order to pinpoint the most impactful ways in which AI can be used and the approaches that are being explored.
Contact: Reihaneh Rabbany (Mila/McGill)
Description: This project aims to study the information posted online surrounding this global pandemic by looking at the mentions of COVID-19 on Twitter. The goal is to discover temporal and spatial trends (to get the pulse and also serve as proxy data), summarize the content (to improve users’ understanding of what is happening) and detect potential polluting groups (to help fight the correlated infodemic). This is important given that the current flattening-of-the-curve models assume some specific degree of adherence from the public, and this is affected by the information they receive. The scale of this infodemic has overwhelmed the content moderators for these platforms (who also have to work from home), which signifies the need for AI solutions.
Contact: Joseph Paul Cohen (Mila/UdeM)
Description: In the context of a COVID-19 pandemic, it is crucial to streamline diagnosis. Data is the first step to developing any diagnostic tool or treatment. While there exist large public datasets of more typical chest X-rays, there is no collection of COVID-19 chest X-rays or CT scans designed to be used for computational analysis. Our team believes that this open database can dramatically improve identification of COVID-19. Notably, this would provide essential data to train and test a deep learning-based system such as Chester, likely using some form of transfer learning. These tools could be developed to identify COVID-19 characteristics compared to other types of pneumonia or in order to predict survival. The database is here and a paper with more information about the project is here. We welcome contributions to the dataset as well as contributors.
Collaborators: Marzyeh Ghassemi (Canada CIFAR AI Chair, Vector Institute, University of Toronto), Joseph Paul Cohen (Mila, Université de Montréal), Chris McIntosh (University of Toronto)
Collaborators: Jian Tang (Canada CIFAR AI Chair, Mila, HEC Montréal), William L. Hamilton (Canada CIFAR AI Chair, Mila, McGill University), Yoshua Bengio (Canada CIFAR AI Chair and co-director, CIFAR Learning in Machines & Brains program, Mila, Université de Montréal), Guy Wolf (Mila, Université de Montréal), Yue Li (Mila, McGill University)
Collaborators: Sarath Chandar (Canada CIFAR AI Chair, Mila, Polytechnique Montréal), Matthew Taylor (Amii, University of Alberta), Sai Krishna (99andBeyond), Karam Thomas (99andBeyond)
Collaborators: Frank Wood (Canada CIFAR AI Chair, Mila, University of British Columbia), Benjamin Bloem-Reddy, Alexandre Bouchard, Trevor Campbell (University of British Columbia)
Mila partnered up with Dialogue and many others to create Chloe: an AI-driven question and answering system. The goal is to help reduce the high call volumes to government helplines at the Quebec Ministry of Health and Social Services. Click here to learn more and here to chat with Chloe