Scaffolding protein functional sites using deep learning

Tuesday April 5th, 4-5pm EST | Doug Tischer & David Juergens, University of Washington

Current approaches to de novo design of proteins harboring a desired binding or catalytic motif require pre-specification of an overall fold and hence considerable trial and error. Here, we describe two complementary approaches to the general functional site design problem. In the first “constrained hallucination” approach, we carry out gradient descent in sequence space to optimize a loss function which simultaneously rewards recapitulation of the desired functional site and the ideality of the surrounding scaffold, supplemented with problem-specific interaction terms. In the second “inpainting” or “missing information recovery” approach, we start from the desired functional site and jointly fill in the “missing” sequence and structure needed to complete the protein in a single forward pass through a modified RoseTTAFold network. We demonstrate the utility of the two methods by addressing a variety of scaffolding design problems, such as epitope scaffolding, metalloproteins and enzymes, and protein binders.

Paper: https://www.biorxiv.org/content/10.1101/2021.11.10.468128v2

Recording link: https://youtu.be/-EJ8SXTBin0