UniMoMo: Unified Generative Modeling of 3D Molecules for De Novo Binder Design

Xiangzhe Kong, Zishen Zhang, Ziting Zhang, Rui Jiao, Jianzhu Ma, Wenbing Huang, Kai Liu, Yang Liu

Abstract:

The design of target-specific molecules such as small molecules, peptides, and antibodies is vital for biological research and drug discovery. Existing generative methods are restricted to single-domain molecules, failing to address versatile therapeutic needs or utilize cross-domain transferability to enhance model performance. In this paper, we introduce **Uni**fied generative **Mo**deling
of 3D **Mo**lecules (UniMoMo), the first framework capable of designing binders of multiple molecular domains using a single model. In particular, UniMoMo unifies the representations of different molecules as graphs of blocks, where each block corresponds to either a standard amino acid or a molecular fragment. Based on these unified representations, UniMoMo utilizes a geometric latent diffusion model for 3D molecular generation, featuring an iterative full-atom autoencoder to compress blocks into latent space points, followed by an E(3)-equivariant diffusion process. Extensive benchmarks across peptides, antibodies, and small molecules demonstrate the superiority of our unified framework over existing domain-specific models, highlighting the benefits of multi-domain training.

Lay Summary:

Finding the right molecule that can stick to (or "bind") a disease-related protein is at the heart of making new medicines. These molecules can come in many forms, like small chemicals, short strings of amino acids (called peptides), or even large immune proteins (called antibodies). Today’s AI tools usually focus on designing just one kind of molecule at a time, with limited flexibility to take advantage of similarities across different types of drugs.

In this work, we introduce **UniMoMo**, a new artificial intelligence model that can design all three major types of drug-like molecules using a single system. The model represents molecules as collections of building blocks (like amino acids or chemical fragments) and learns how these blocks interact with target proteins. We use advanced AI techniques to first compress the detailed atomic structure into a simpler form, then generate new molecules in this compressed space, and finally rebuild the full 3D atomic structure.

We tested UniMoMo on a wide range of benchmarks and showed that it performs better than specialized models built for only one type of molecule. We also showed that the model can generalize—for example, designing small molecules that borrow helpful patterns from peptide or antibody interactions. This flexibility brings us a step closer to using a single tool to design many kinds of therapeutics, potentially speeding up and simplifying drug discovery.

Link To Code: https://github.com/kxz18/UniMoMo

Primary Area: Applications->Health / Medicine

Keywords: Unified Generative Framework, Molecular Binder Design, Drug Discovery, Geometric Latent Diffusion

Submission Number: 6355