An All-Atom Generative Model for Designing Protein Complexes

2025-04-17

AI Summary

Modeling multi-chain protein complexes remains a critical bottleneck in structural biology and protein design, as existing single-chain models (e.g., ESM, AlphaFold) lack the capacity to capture atomic-level inter-chain interactions. To address this, we introduce APM—the first end-to-end, all-atom generative model for multi-chain proteins—integrating geometrically invariant representations, inter-chain attention mechanisms, and a deep generative architecture that supports both zero-shot sampling and supervised fine-tuning. APM achieves state-of-the-art performance across multi-chain folding, inverse folding, and *de novo* binding interface design. It generates high-fidelity complex structures, enables function-driven binding affinity engineering, and produces experimentally verifiable outputs. By unifying geometric reasoning with scalable generative learning, APM establishes a new paradigm for rational, multi-chain protein design.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Multiagent Systems: Mechanism Design
• Natural Language Processing: Generation

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Economics, Online Markets and Human Computation: Economic ramifications for generative AI infrastructure and applications
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Proteins typically exist in complexes, interacting with other proteins or biomolecules to perform their specific biological roles. Research on single-chain protein modeling has been extensively and deeply explored, with advancements seen in models like the series of ESM and AlphaFold. Despite these developments, the study and modeling of multi-chain proteins remain largely uncharted, though they are vital for understanding biological functions. Recognizing the importance of these interactions, we introduce APM (All-Atom Protein Generative Model), a model specifically designed for modeling multi-chain proteins. By integrating atom-level information and leveraging data on multi-chain proteins, APM is capable of precisely modeling inter-chain interactions and designing protein complexes with binding capabilities from scratch. It also performs folding and inverse-folding tasks for multi-chain proteins. Moreover, APM demonstrates versatility in downstream applications: it achieves enhanced performance through supervised fine-tuning (SFT) while also supporting zero-shot sampling in certain tasks, achieving state-of-the-art results. Code will be released at https://github.com/bytedance/apm.

Problem

• Modeling multi-chain protein interactions for biological understanding
• Designing protein complexes with binding capabilities from scratch
• Performing folding and inverse-folding tasks for multi-chain proteins

Innovation

• Generative model for multi-chain protein complexes
• Atom-level modeling of inter-chain interactions
• Supports folding and inverse-folding tasks