Presented on October 22nd 2025 by Gabe Rocklin
http://www.rocklinlab.org/
abstract: We developed new multiplexed approaches for measuring protein folding stability and aggregation propensity at scale, enabling us to measure folding for ~3 million protein domains and stress-induced aggregation for ~20,000 domains. Fine-tuned protein language models trained on each dataset successfully predict ~80% of the variance in folding stability and ~50% of the variance in protein aggregation, opening many exciting applications in protein engineering.