Center for Biomolecular Structure Lecture Series
"CBMS Lecture Series - Ivan Anishchenko"
Presented by Ivan Anishchenko, University of Washington
Wednesday, January 19, 2022, 1:30 pm — Videoconference / Virtual Event (see link below)
Deep neural networks for protein-protein complex prediction and design
AlphaFold and RoseTTAfold now enable the very accurate prediction of the structures of protein monomers from sequence information alone. Generating models of protein-protein complexes is more challenging however, as these methods rely in part on evolutionary covariation. To overcome this challenge, we developed a method for systematically pairing orthologs from different organisms in paired multiple sequence alignments and used a combination of AlphaFold and RoseTTAFold to identify and build accurate models of core eukaryotic protein complexes within the S.cerevisiae proteome, including 106 previously unidentified assemblies and 806 that have not been structurally characterized. On the protein design side, I will present how deep networks trained to predict native protein structures from their sequences can be inverted to create new proteins with sequences unrelated to those of native proteins and that fold into stable structures. Experimental studies on 129 of network-generated proteins showed that many of them fold into the designed structures. The developed network-based protein design approach can be readily generalized to include specific structural features, like binding motifs or catalytic sites, around which the network can generate new protein inhibitors or enzyme catalysts.
Hosted by: Vivian Stojanoff
17208 | INT/EXT | Events Calendar