Regina Barzilay Group
AI For Healthcare and Life Sciences
Current Research Areas
Modeling molecules and their interactions
Understanding how proteins and small molecules interact with one another is a component in improving our understanding of many biological processes and speeding up drug discovery. In particular, molecular docking (i.e. finding the 3D structure of these interactions), holds the keys to predicting interaction strength and how this can be altered. Our lab has pioneered the use of machine learning methods to generate the structure of the binding poses, paving the way for the replacement of costly and inaccurate traditional search-based methods.
We are also developing methods for modeling molecular interactions in the context of cellular metabolism and immunology. Our group has developed methods for metabolism ranging from from the individual reaction level, which is vital to enzyme screening and drug discovery, to the overall metabolic system, which is key to understanding disease and optimizing biomanufacturing. We are also utilizing binding models to help increase our understanding of immune system function.
Selected publications:
Independent SE(3)-Equivariant Models for End-to-End Rigid Protein Docking
EquiBind: Geometric Deep Learning for Drug Binding Structure Prediction
DiffDock: Diffusion Steps, Twists, and Turns for Molecular Docking
Biomolecule generation and design
Our research advances machine learning techniques to model and design novel biomolecules. We work closely with biologists and chemists to tie machine learning research with real world impact. Our interests span many problems including generative modeling, adaptive experimental design, low-N optimization, and inverse design.
Selected publications:
SE(3) diffusion model with application to protein backbone generation
De novo design of protein structure and function with RFdiffusion
Optimizing protein fitness using Gibbs sampling with Graph-based Smoothing
Clinical Machine Learning
We develop methods to leverage machine learning to enhance clinical care, and our efforts encompass the entire patient journey, from pre-diagnosis to post-treatment. Our research focuses on improving health outcomes for a broad spectrum of conditions, including cancer, blood diseases, organ transplants, and diabetes, among others. In order to bridge the gap between the research world and real-world clinical practice, we focus on developing tools that facilitate the practical implementation of AI discoveries. This covers various aspects such as clinical decision making from AI predictions, statistical guarantees on these predictions, and bias detection and prevention. Additionally, we collaborate closely with physicians to design user-friendly frameworks, increasing the likelihood of adoption by other healthcare institutions.
Selected publications:
Optimizing risk-based breast cancer screening policies with reinforcement learning
A deep learning mammography-based model for improved breast cancer risk prediction.