Predicting Small-Molecule Binding Sites Using AlphaFold2 and Leveraging Deep Learning Model Embeddings for Protein Property Prediction

Explore cutting-edge computational approaches to protein analysis and drug discovery through this comprehensive seminar from the Broad Institute's Models, Inference and Algorithms series. Learn about AF2BIND, a novel logistic regression model that leverages AlphaFold2's pair representation features to predict small-molecule binding sites in proteins without relying on homology modeling, multiple sequence alignments, or prior knowledge of pocket-compatible ligands. Discover how this de novo approach identifies binding sites across the human proteome, creating a database of thousands of previously unseen binding sites in disease-relevant proteins, and understand how interpretable model features can predict chemical properties of compatible ligands for focused drug discovery efforts. Gain foundational knowledge through the primer session on leveraging deep learning model embeddings for protein property prediction and design, comparing embedding generation methods in graph neural networks like AlphaFold2 and ProteinMPNN versus protein language models, and understanding how these rich biochemical, evolutionary, and biophysical representations enable accurate protein property prediction essential for comprehending the AF2Bind model's functionality.