Manipulating 3D Molecules in a Fixed-Dimensional SE(3)-Equivariant Latent Space

2025-06-01

AI Summary

Current drug discovery urgently requires flexible, unsupervised 3D-structure-aware molecular editing methods that avoid task-specific fine-tuning. To address this, we propose the first zero-shot 3D molecular manipulation framework, which constructs a fixed-dimensional, SE(3)-equivariant shared latent space—thereby decoupling representation from atom count—to enable simultaneous atomic-number editing, conformational reconstruction, and multi-attribute interpolation. Our approach integrates an SE(3)-equivariant graph neural network, the MolFLAE variational autoencoder, and a Bayesian flow network (BFN). On unconditional 3D molecular generation, it achieves state-of-the-art performance. Crucially, in a real-world application to glucocorticoid receptor ligand optimization, the method significantly improved hydrophilicity while fully preserving critical binding interactions—demonstrating both practical efficacy and strong generalization in drug design.

Technology Category

• Machine Learning: Deep Generative Models & Autoencoders
• Computer Vision: 3D Computer Vision
• Search and Optimization: Learning to Search

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Medicinal chemists often optimize drugs considering their 3D structures and designing structurally distinct molecules that retain key features, such as shapes, pharmacophores, or chemical properties. Previous deep learning approaches address this through supervised tasks like molecule inpainting or property-guided optimization. In this work, we propose a flexible zero-shot molecule manipulation method by navigating in a shared latent space of 3D molecules. We introduce a Variational AutoEncoder (VAE) for 3D molecules, named MolFLAE, which learns a fixed-dimensional, SE(3)-equivariant latent space independent of atom counts. MolFLAE encodes 3D molecules using an SE(3)-equivariant neural network into fixed number of latent nodes, distinguished by learned embeddings. The latent space is regularized, and molecular structures are reconstructed via a Bayesian Flow Network (BFN) conditioned on the encoder's latent output. MolFLAE achieves competitive performance on standard unconditional 3D molecule generation benchmarks. Moreover, the latent space of MolFLAE enables zero-shot molecule manipulation, including atom number editing, structure reconstruction, and coordinated latent interpolation for both structure and properties. We further demonstrate our approach on a drug optimization task for the human glucocorticoid receptor, generating molecules with improved hydrophilicity while preserving key interactions, under computational evaluations. These results highlight the flexibility, robustness, and real-world utility of our method, opening new avenues for molecule editing and optimization.

Problem

• Optimizing 3D drug structures while retaining key features
• Learning fixed-dimensional SE(3)-equivariant latent space for molecules
• Enabling zero-shot manipulation of molecular properties and structures

Innovation

• SE(3)-equivariant VAE for 3D molecules
• Fixed-dimensional latent space navigation
• Zero-shot manipulation via Bayesian Flow Network