AI-guided design of de novo proteins: From new structures to programmable dynamics, protein networks, and cellular functions

Methods from artificial intelligence (AI) trained on large datasets of sequences and structures can now “write” proteins with new shapes and molecular functions de novo, without starting from proteins found in nature [1]. Many protein functions - those critical for signal processing, energy conversion, and multistep catalytic transformations - require defined structural transitions between dynamically interconverting protein conformations. These functions are currently challenging to design de novo even with advanced AI models that are trained to predict and design static structures. I will describe our recent progress with (i) designing dynamic proteins [2] and (ii) developing advanced AI methods that can generate proteins with multiple conformational states and physically realistic paths between them. I will show recent results demonstrating the design of diverse, de novo protein conformational changes, suggesting that protein motions might be more designable than previously thought. Since suitable and large-scale experimental data on protein dynamics remain a critical bottleneck for developing improved models, I will describe how we are designing and experimentally characterizing hundreds of predicted protein conformational switches in high-throughput. Finally, I will discuss advances moving from design of bimolecular recognition to design of reaction networks. Exciting frontiers lie in constructing synthetic cellular signaling from the ground up using programmable de novo proteins [1].

Speaker: Tanja Kortemme, UC, San Francisco