How, and What, Does AlphaFold Learn About Protein Structure?

DeepMind’s AlphaFold transformed structural biology, offering an unprecedented way to predict the three-dimensional structure of proteins directly from primary amino acid sequences. While AlphaFold performs exceedingly well at its task, questions remain over the precise nature of the internal representation it builds.

In this Presidential Lecture, AlQuraishi will discuss recent evidence on the extent to which AlphaFold appears to learn to perform implicit physical modeling, and how this knowledge is acquired during its training. He will present use cases in which this implicit knowledge unlocks new capabilities and where it falls short. He also plans to discuss the differences between AlphaFold 2 and AlphaFold 3 regarding this question of implicit physical knowledge, and how their architectural differences may have led to different inductive priors.

AlQuraishi s an assistant professor in the Department of Systems Biology at Columbia University and a member of Columbia’s Program for Mathematical Genomics, where he works at the intersection of machine learning, biophysics and systems biology. AlQuraishi earned an MS in statistics and a Ph.D. in genetics from Stanford University. He subsequently joined the Department of Systems Biology at Harvard Medical School as a departmental fellow and a fellow in systems pharmacology, where he developed the first end-to-end differentiable model for learning protein structure from data. Prior to starting his academic career, AlQuraishi spent three years founding two startups in the mobile computing space.