All-atom representations improve unified molecular binder design models
Towards Generalizing Pretrained All-Atom Representations to Unified Binder Design
Machine LearningArtificial Intelligence
Summary
Designing molecules that bind to targets like proteins or peptides is important for medicine and biology. Existing models often focus on a specific type of molecule using detailed atomic information, but struggle to work across different types together. The authors created PairFuse, a method that combines detailed atomic models with a versatile shape-based model to design various molecular binders in one system. This approach improves performance in designing antibody, peptide, and small molecule binders by better integrating atomic details and structural relationships.
What this means in practice
- •For biotech engineers: Design unified molecular binders that work across proteins, peptides, and small molecules with improved accuracy by integrating atomic and structural features.
- •For pharmaceutical developers: Enhance drug candidate design for complex targets by leveraging a single model that represents diverse molecular modalities with detailed atomic information.
Authors
Maojiang Su, Jiale Chen, Qijie Zhu, Shang Wu, Jiayi Wang, Jihai Zhao, Guo Ye, Yibo Wen, Han Liu
Abstract
PairFormer-based all-atom models demonstrate strong capabilities in biomolecular binder design. However, existing methods often use all-atom representations for modality-specific binder design, such as protein and peptide design. In contrast, recent geometric latent diffusion models pursue unified binder design by representing molecular blocks from different modalities in a shared latent space. These complementary strengths motivate us to investigate how to generalize the use of pretrained all-atom representations to unified binder design. We introduce PairFuse, a parameter-efficient method that enables this generalization through representation fusion. Specifically, we use lightweight layerwise adapters to integrate pretrained all-atom representations into an independently pretrained geometric latent diffusion model. Our adapters integrate pretrained all-atom representations through two mechanisms: enriching the diffusion model's molecular block features and incorporating pairwise structural information into the geometric diffusion model's denoiser. Theoretically, we prove that the augmented denoiser preserves geometry and pairwise attention fusion provides additional attention expressivity beyond feature-only fusion. Experimentally, across antibody, peptide, and small molecule binder-design benchmarks, PairFuse achieves stronger performance than the pretrained diffusion model and existing representation-fusion baselines.