Pocket2Mol: Efficient Molecular Sampling Based on 3D Protein Pockets

2022-05-15

International Conference on Machine Learning

AI Summary

To address 3D drug molecule generation under geometric constraints imposed by protein binding pockets, this paper proposes the first end-to-end differentiable 3D conditional molecular generation framework. Methodologically, it introduces an E(3)-equivariant dual-module architecture: a graph neural network (GNN) encoder that jointly embeds pocket atomic coordinates and ligand chemical graph structure, and an explicit probability distribution sampler—bypassing MCMC—that jointly models atomic 3D positions, bond types, and functional groups. Its key innovation lies in the first-ever E(3)-equivariant joint modeling of both pocket geometry and molecular chemical details, enabling efficient, deterministic 3D molecular sampling. Experiments demonstrate that generated molecules significantly outperform existing baselines in binding affinity (predicted via docking and scoring), drug-likeness (assessed by QED and SA scores), and synthetic accessibility—thereby substantially improving success rates in structure-based drug design.

Abstract

Deep generative models have achieved tremendous success in designing novel drug molecules in recent years. A new thread of works have shown the great potential in advancing the specificity and success rate of in silico drug design by considering the structure of protein pockets. This setting posts fundamental computational challenges in sampling new chemical compounds that could satisfy multiple geometrical constraints imposed by pockets. Previous sampling algorithms either sample in the graph space or only consider the 3D coordinates of atoms while ignoring other detailed chemical structures such as bond types and functional groups. To address the challenge, we develop Pocket2Mol, an E(3)-equivariant generative network composed of two modules: 1) a new graph neural network capturing both spatial and bonding relationships between atoms of the binding pockets and 2) a new efficient algorithm which samples new drug candidates conditioned on the pocket representations from a tractable distribution without relying on MCMC. Experimental results demonstrate that molecules sampled from Pocket2Mol achieve significantly better binding affinity and other drug properties such as druglikeness and synthetic accessibility.

Problem

• Sampling molecules meeting protein pocket constraints
• Incorporating spatial and bonding atom relationships
• Improving drug binding affinity and properties

Innovation

• E(3)-equivariant generative network for molecular sampling
• Graph neural network capturing spatial and bonding relationships
• Efficient algorithm sampling drug candidates without MCMC