Equivariant Diffusion for Molecule Generation in 3D

TL;DR

The paper presents an E(3)-equivariant diffusion model (EDM) for 3D molecule generation, enabling generation that respects Euclidean symmetries by jointly denoising both continuous atom coordinates and discrete atom types. It introduces a probabilistic analysis that allows likelihood computation within the diffusion framework, facilitating principled evaluation. Empirically, EDM achieves superior sample quality and training efficiency compared to prior 3D molecular generative methods, highlighting its practicality for scalable 3D molecular design. This approach advances geometry-aware generative modeling with potential impacts on drug discovery and materials science by producing physically plausible 3D molecular structures with symmetry-consistent representations.

Abstract

This work introduces a diffusion model for molecule generation in 3D that is equivariant to Euclidean transformations. Our E(3) Equivariant Diffusion Model (EDM) learns to denoise a diffusion process with an equivariant network that jointly operates on both continuous (atom coordinates) and categorical features (atom types). In addition, we provide a probabilistic analysis which admits likelihood computation of molecules using our model. Experimentally, the proposed method significantly outperforms previous 3D molecular generative methods regarding the quality of generated samples and efficiency at training time.

Figures (1)

• Figure 1: Historical locations and number of accepted papers for International Machine Learning Conferences (ICML 1993 -- ICML 2008) and International Workshops on Machine Learning (ML 1988 -- ML 1992). At the time this figure was produced, the number of accepted papers for ICML 2008 was unknown and instead estimated.