9I制作厂免费

Understanding and guiding evolution with protein language models

Peter S. Kim, Stanford University
Tuesday January 20, 12-1pm
Zoom Link:听
In Person: 550 Sherbrooke, Room 189

Abstract: Nature is the most powerful protein engineer, and evolution is nature鈥檚 design process. We have been using protein language models to better understand the rules of natural protein evolution. By reconstructing a global landscape of protein evolution through local evolutionary predictions that we refer to as evolutionary velocity (evo-velocity), we can predict evolution in diverse landscapes, including the progression of viral outbreaks, and the evolution of eukaryotic proteins over geologic eons. The same ideas can also be used to evolve natural proteins with improved fitness, in a highly efficient manner. To improve predictive capabilities, we incorporate structural information with a structure-informed language model. Importantly, the use of only protein backbone coordinates is sufficient to learn principles of binding that generalize to protein complexes and enable antibody engineering with unprecedented efficiency. Together, these results lay the groundwork for more potent and resilient therapeutic design in unsupervised settings and in the absence of task-specific training data.