Event Overview
Predicting the structures of biomolecular complexes requires models that can handle longer sequences while managing substantial computational and memory demands. This webinar explores long-context protein folding with ESMFold2, from biological motivation and existing limitations to approaches for extending its reach. Researchers from Biohub, Anthropic, and NVIDIA will cover Biohub’s internal optimizations, introduce NVIDIA BioNeMo Fold-CP and its use of context parallelism to distribute folding workloads across GPUs, and discuss Anthropic’s latest advances in ESMFold2, including speedups, accuracy comparisons, and a demonstration with large protein complexes. We’ll also compare supported capabilities, hardware requirements, and practical constraints to help participants understand when to use each approach.
Speakers
Allan dos Santos Costa
Allan dos Santos Costa is a researcher at Biohub working on deep learning for structural biology. He holds a Ph.D. from MIT, where he investigated generative physics-inspired models for biomolecules.
Simon Chu
Simon Chu works as a research scientist on the NVIDIA BioNeMo team. With a Ph.D. in biophysics, he focuses on AI models for protein sequence modeling, structure prediction, and design.
Dejun Lin
Dejun Lin is an engineer at NVIDIA. He primarily works on accelerating AI models for biological and drug discovery applications. With a Ph.D. in Biophysics, he has extensive experience in high-performance computing (HPC) with a deep passion for AI and machine learning models.
Richard Shuai
Richard Shuai is a Member of Technical Staff at Anthropic. He conducted his Biophysics Ph.D. research in Po-Ssu Huang’s lab at Stanford University, where he focused on generative models for protein design.Presented in Collaboration With
