Unified all-atom molecule generation with neural fields

2025-11-19

AI Summary

Current generative models for structure-based drug design suffer from modality specificity, limiting their ability to jointly model multi-scale molecular structures. To address this, we propose the first modality-agnostic, all-atom neural field generation framework: molecules are represented as continuous 3D atomic density fields, integrated with fractional diffusion and 3D vision architectures (e.g., ViT, U-Net). This formulation naturally supports variable-length sequences, non-canonical residues, and cross-scale molecular entities—including small molecules, macrocyclic peptides, and antibody CDRs. Our work establishes the first unified integration of neural fields and computer vision paradigms for all-atom molecular generation. Experiments demonstrate state-of-the-art performance across diverse in silico generation tasks. Notably, the framework successfully de novo designs novel antibody binders validated by in vitro binding assays.

Technology Category

• Machine Learning: Deep Neural Architectures and Foundation Models
• Computer Vision: Diffusion Models for Vision
• Natural Language Processing: Generation

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Social Networks and Social Media: Generative AI / large language models and their impact on social systems

Abstract

Generative models for structure-based drug design are often limited to a specific modality, restricting their broader applicability. To address this challenge, we introduce FuncBind, a framework based on computer vision to generate target-conditioned, all-atom molecules across atomic systems. FuncBind uses neural fields to represent molecules as continuous atomic densities and employs score-based generative models with modern architectures adapted from the computer vision literature. This modality-agnostic representation allows a single unified model to be trained on diverse atomic systems, from small to large molecules, and handle variable atom/residue counts, including non-canonical amino acids. FuncBind achieves competitive in silico performance in generating small molecules, macrocyclic peptides, and antibody complementarity-determining region loops, conditioned on target structures. FuncBind also generated in vitro novel antibody binders via de novo redesign of the complementarity-determining region H3 loop of two chosen co-crystal structures. As a final contribution, we introduce a new dataset and benchmark for structure-conditioned macrocyclic peptide generation. The code is available at https://github.com/prescient-design/funcbind.

Problem

• Generating diverse all-atom molecules across atomic systems for drug design
• Overcoming modality limitations in structure-based molecular generation
• Creating target-conditioned molecules from small compounds to antibodies

Innovation

• Neural fields represent molecules as continuous densities
• Score-based generative models adapted from computer vision
• Unified model handles diverse atomic systems and molecules