Learning from natural antibodies for sequence generation and fast structure prediction

Presented on March 2nd, 2022 by Jeff Ruffolo. Hosted by Chris Bahl and Sergey Ovchinnikov.
Abstract:
Billions of natural antibody sequences have been identified through immune repertoire sequencing studies. Using this data, we developed a novel infilling language model for antibody sequence generation and targeted diversification. Pairing this data with structural information, we trained an end-to-end model for antibody structure prediction, which enables accurate structural modeling at scale.