Predicting Equilibrium Distributions for Molecular Systems with Deep Learning

Explore a comprehensive lecture on predicting equilibrium distributions for molecular systems using deep learning. Delve into traditional methods for obtaining distributions and learn how AI can accelerate sampling processes. Discover the capabilities of Distributional Graphormer (DiG), a novel deep learning framework that transforms simple distributions towards equilibrium distributions. Examine topics such as diffusion models, equivariant Graphormer, training from energy functions and simulation data, protein conformation sampling, and inverse design. Gain insights into applications including ligand structure sampling, catalyst-adsorbate sampling, and property-guided structure generation. Understand how this advancement in methodology opens new research opportunities in molecular science and drug discovery.

- Intro
- Obtaining Distribution: Traditional Methods
- Using AI to Accelerate Sampling
- Diffusion Models
- Distributional Graphormer: Capabilities
- Equivariant Graphormer
- Training From Energy Function & Simulation Data
- Protein Conformation Sampling
- Sampling Metastables
- Conformation Transition Pathway Prediction
- Protein-Ligand Binding Sampling
- Catalyst Absorption Sampling
- Density Estimation
- Inverse Design
- Q+A