Backbone Generation with RFdiffusion

This video covers RFdiffusion for protein backbone generation, a machine learning denoising approach finetuned from RoseTTAFold, a structure prediction network. After a short overview of the model’s fundamentals, Nick describes inputs and outputs, architecture, loss calculations, versions of RFdiffusion, and limitations and extensions. Video from the Rosetta Commons ML Bootcamp (October 2024) Video Instructor: Nick Randolph (UNC Chapel Hill) Credits: Course Instructor: Nick Randolph Course Teaching Assistants: Fatima Hitawala, Amrita Nallathambi, Ben Orr RC Leadership and NSF Sponsored Grant PIs: Julia Koehler Leman & Jeffrey Gray RC Education Director: Ashley Vater Videographer: Ryan Caster Rosetta Bootcamp Trainee Participants. Funding: Rosetta Commons and National Science Foundation 00:00 - Intro 00:10 - RFDiffusion 08:08 - Inputs and Outputs 12:41 - RFDiffusion Overview 13:32 - RoseTTAFold2 Architecture 15:43 - RoseTTAFold and RFDiffusion Losses 18:27 - Versions of RFDiffusion 20:21 - Potentials 22:33 - Limitations 23:06 - Extensions