Explore a 56-minute conference talk on PepFlow, a groundbreaking deep learning approach for direct conformational sampling from peptide energy landscapes. Delve into the innovative use of generalized Boltzmann generators, modularized generation processes, and hypernetworks to overcome challenges in all-atom modeling of peptides. Learn how PepFlow accurately predicts peptide structures, recapitulates experimental ensembles efficiently, and samples conformations with specific constraints like macrocyclization. Discover the potential impact of this research on drug discovery and biomolecular structure prediction through detailed explanations of the methodology, performance on SLiMs, and generalization capabilities for macrocyclic conformations.
Overview
Syllabus
- Intro
- Boltzmann Generators
- Modularizing Conformation Generation
- Increasing Effectivity with a Hypernetwork
- Summary of the Pepflow Approach
- Training by Energy to Improve Concordance of Backbones
- Performance of Pepflow on SLiMs
- Can Pepflow Generalize to the Sampling of Macrocyclic Conformations?
- Conclusions
- Q+A
Taught by
Valence Labs
