Bindcraft

This video covers Bindcraft, a one-shot design of functional binders. Starting with a brief overview of current binder design tools and conceptually explaining the Bindcraft’s central hallucination process that engages loss and confidence metrics to predict sequence from structure. Cho then delves into the internal workflow of Bindcraft describing AF2 Mulitmer for backbone and sequence codesign, sequence optimization with MPNNsol M, and validation/filtering with AF2Monomer. The successes and limitations of this tool are addressed. Video from the Rosetta Commons PPI Workshop (February 2025) Video Instructor: Yehlin Cho (Massachusetts Institute of Technology, MIT) Credits: Instructor: Yehlin Cho Teaching Assistants: Yehlin Cho, Cyrus Haas, and Amrita Nallathambi RC Leadership and NSF Sponsor Grant PIs: Julia Koehler Leman & Jeffrey Gray RC Education Director: Ashley Vater Videographer: Canyon Florey Rosetta Workshop Participants 00:00 - Introduction 00:26 - Bindcraft 00:38 - State of computational binder design before DL 02:41 - AlphaFold2 03:31 - Confidence metrics 09:09 - Previous work 1: Designing a new sequence given a structure 10:07 - Previous work 2: Designing a new sequence given a structure 10:42 - Bindcraft: Designing a binder given a target 11:30 - AF2 Hallucinations 12:10 - Four Stages 15:25 - Structure variations 17:37 - ProteinMPNN Sequence Design 18:56 - AF2 Structure Prediction 20:22 - Limitations