Unified all-atom molecule generation with neural fields
Matthieu Kirchmeyer, Pedro O. Pinheiro, Emma Willett, Karolis Martinkus, Joseph Kleinhenz, Emily K. Makowski, Andrew M. Watkins, Vladimir Gligorijevic, Richard Bonneau, Saeed Saremi
2025-11-26
Summary
This research introduces FuncBind, a new computer program designed to create molecules for potential use as drugs. It's special because it can design molecules of different types and sizes, all while considering the specific protein they need to interact with.
What's the problem?
Existing methods for designing drugs using computers often focus on only one type of molecule. This limits their usefulness because many drugs aren't simple, and researchers need tools that can handle a wider variety of molecular structures. It's hard to create a single program that can effectively design both small molecules and large, complex ones like proteins or antibodies.
What's the solution?
The researchers used techniques from computer vision – the field that allows computers to 'see' and interpret images – to represent molecules in a new way. Instead of thinking of molecules as collections of individual atoms, they treated them as continuous densities. They then used a type of artificial intelligence called a 'score-based generative model' to create new molecules that fit a desired target structure. This approach allows FuncBind to work with different kinds of molecules and even molecules containing unusual building blocks.
Why it matters?
FuncBind is important because it provides a more versatile tool for drug discovery. It can generate potential drug candidates across a broader range of molecular types, including complex structures like antibodies. The researchers also created a new dataset to help others test and improve similar programs, and even successfully used FuncBind to design new antibody binders in the lab, showing its real-world potential.
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.