VecMol: Vector-Field Representations for 3D Molecule Generation

Yuchen Hua, Xingang Peng, Jianzhu Ma, Muhan Zhang

TL;DR: We generate 3D molecules by learning and diffusing continuous vector fields instead of graphs, simplifying geometry–chemistry modeling while achieving competitive results on QM9 and GEOM-Drugs.

Abstract: Generative modeling of three-dimensional (3D) molecules is a fundamental yet challenging problem in drug discovery and materials science. Existing approaches typically represent molecules as 3D graphs and co-generate discrete atom types with continuous atomic coordinates, leading to intrinsic learning difficulties such as heterogeneous modality entanglement and geometry–chemistry coherence constraints. We propose VecMol, a novel representation that models 3D molecules as continuous vector fields over Euclidean space, where vectors point toward nearby atoms and implicitly encode molecular structure. The vector field is parameterized by a neural field and generated using a latent diffusion model, avoiding explicit graph generation and decoupling structure learning from discrete atom instantiation. Experiments on the QM9 and GEOM-Drugs benchmarks demonstrate that VecMol achieves competitive generation quality, suggesting vector-field-based representations as a promising new direction for 3D molecular generation.

Lay Summary:

Discovering new drugs often requires designing molecules with precise three-dimensional shapes, but predicting these shapes is surprisingly difficult. Existing AI methods typically represent molecules as disconnected lists of atoms or low-resolution 3D grids, which either limits the complexity of molecules they can handle or demands excessive computing power.

VecMol introduces a fundamentally different approach. Instead of listing atoms, it represents a molecule as a continuous "direction map" over 3D space — imagine being at any point in space and an arrow tells you which way to go to find the nearest atom. An AI model learns to generate these direction maps from scratch, and the atoms are then recovered by following the arrows.

VecMol works for both small molecules (dozens of atoms) and large drug-like molecules (hundreds of atoms), matching or exceeding the performance of current best methods. By rethinking how molecules are represented, this work opens a more efficient path for AI-driven drug discovery.

Link To Code: https://github. com/MuLabPKU/VecMol

Primary Area: Applications->Health / Medicine

Keywords: Neural fields, 3D molecular generation, Diffusion models, Equivariant neural networks, Molecular representation learning

Submission Number: 9854