Protein-SE(3): Unified Framework and Comprehensive Benchmark for SE(3)-based Protein Structure Design

ICLR 2026 Conference Submission15753 Authors

Keywords: Protein Structure Design, Unified Training Framework, Mathematical Decoupling, Comprehensive Benchmark

Abstract: SE(3)-based generative models have shown great promise in protein geometry modeling and effective structure design. However, the field currently lacks a pipeline to support consistent re-training and fair comparison across different methods. In this paper, we propose Protein-SE(3), a unified framework accompanied by the comprehensive benchmark for SE(3)-based protein design. Protein-SE(3) integrates recent advanced methods, supports diverse evaluation metrics and also develops a mathematical decoupling toolkit. Specifically, state-of-the-art generative models for typical protein design tasks (unconditional generation and motif scaffolding), from multiple perspectives like DDPM (Genie1 and Genie2), Score Matching (FrameDiff and RfDiffusion) and Flow Matching (FoldFlow and FrameFlow) are systematically incorporated into our framework. All methods are re-trained on identical datasets and evaluated with consistent metrics, ensuring fair and reproducible comparison. Furthermore, our proposed decoupling toolkit abstracts the mathematical foundations of generative models, facilitating rapid prototyping of future algorithms without reliance on explicit protein structures. Taken together, Protein-SE(3) establishes a standardized foundation for the advancing research field of SE(3)-based protein design.

Primary Area: datasets and benchmarks

Submission Number: 15753