ProtMD: Incorporating Conformation Flexibility for Drug Binding via Pretraining on MD Simulations

Explore a 44-minute conference talk on ProtMD, a novel approach incorporating conformation flexibility for drug binding through pretraining on molecular dynamics simulations. Delve into the limitations of the "lock-and-key" theory and discover how ProtMD leverages spatial-temporal pre-training methods to capture protein flexibility. Learn about self-supervised learning tasks, including atom-level prompt-based denoising and conformation-level snapshot ordering, designed to extract valuable information from MD trajectories. Examine the impressive performance improvements in binding affinity prediction and ligand efficacy assessment. Gain insights into the correlation between conformation motion and ligand binding strength. Follow along as the speaker covers motivations, molecular dynamics simulations, protein self-supervised learning, model overview, snapshot ordering, graph matching networks, pre-training datasets, experimental results, and the underlying mechanisms of ProtMD's effectiveness.

- Intro
- Motivations
- Molecular Dynamics Simulations
- Protein Self-Supervised Learning
- Model Overview
- Snapshot Ordering
- Graph Matching Network
- Pre-raining Dataset & Experimental Results
- How & Why Does ProtMD work?
- Q+A and Discussion