Reproduce “SymFold: Synergizing Evolutionary and Structural Priors for Accurate Protein Inverse Folding” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Handong Wang, Jiaxin Qi, Baisheng Lai, Jianqiang Huang
SymFold, developed by researchers at the Chinese Academy of Sciences, introduces a symmetric dual-path architecture that synergizes evolutionary knowledge from protein language models (PLMs) and structural priors from multimodal protein language models (MPLMs) for protein inverse folding. The framework achieved state-of-the-art sequence recovery rates, reaching 63.11% on CATH4.2 and outperforming previous PLM-based methods by 2.34 percentage points, while also demonstrating improved generalization to novel structures and better biological plausibility of designed sequences.
Key insights
- Synergistically combining both evolutionary (PLM) and structural (MPLM) priors is crucial for optimal protein inverse folding performance, as neither source alone provides a complete understanding of sequence-structure relationships.
- Dynamically weighting the influence of structural cues and sequence dependencies at a per-residue level enhances refinement adaptability, reflecting the varying biochemical requirements across different amino acid positions.
- Aligning the training process with iterative inference through self-correction effectively mitigates exposure bias, contributing to the model's overall stability and performance during multi-step sequence prediction.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.