Evolutionary scale language models
Alexander Rives (Meta)
Colloquium
Thursday, April 20, 2023, 3:30 pm
Abstract
Generative artificial intelligence has the potential to enable programmable biology. Advanced protein language models demonstrate emergent learning of atomic resolution structure and protein design principles. We study the scaling of language models across orders of magnitude in the number of parameters, and observe the development of an atomic level structure prediction capability. When applied generatively language models generalize beyond natural proteins. We have evaluated AI generated proteins in the laboratory with high success rates, including for proteins without significant sequence similarity to natural proteins. These developments enable programmable design of de novo protein sequences and structures of high complexity.
Bio
Alexander Rives is the Science Lead for the protein team and Research Scientist at Meta Fundamental AI Research (FAIR), and Ph.D. candidate at NYU.