SymFold: Synergizing Evolutionary and Structural Priors for Accurate Protein Inverse Folding
A paper titled 'SymFold: Synergizing Evolutionary and Structural Priors for Accurate Protein Inverse Folding' was published on arXiv (cs.AI) on 2026-09-01. It introduces a symmetric dual-path architecture that combines protein language models (PLMs) for sequence evolution knowledge and multimodal protein language models (MPLMs) for structural knowledge to iteratively guide protein sequence generation. The method achieves state-of-the-art performance on standard protein inverse folding benchmarks, surpassing prior approaches.
Protein inverse folding aims to recover amino acid sequences from a given 3D protein structure, with applications in enzyme engineering and drug discovery. Existing methods often use a serial pipeline where a structure encoder predicts a coarse sequence refined by PLMs, but refinement is limited by upstream quality. MPLMs can directly encode structure but are ineffective for inverse folding. SymFold proposes a symmetric dual-path architecture leveraging both PLMs and MPLMs to iteratively guide sequence generation. Experiments show state-of-the-art performance on standard benchmarks, and ablation studies validate the symmetric design.
The key technical contribution is a symmetric dual-path architecture that integrates pretrained sequence evolution knowledge from PLMs and pretrained structural knowledge from MPLMs in an iterative generation process, overcoming limitations of serial pipelines and direct MPLM encoding. The approach achieves state-of-the-art results on protein inverse folding benchmarks, indicating effective fusion of evolutionary and structural priors.
Improved protein inverse folding can accelerate enzyme engineering and drug discovery by enabling more accurate sequence design for desired structures. This research may influence the development of AI tools for protein design in biotech and pharmaceutical industries.
The method could reduce costs and time in protein engineering by improving sequence prediction accuracy, potentially leading to new therapeutics and industrial enzymes. It may be valuable for companies developing AI-driven protein design platforms.
Next signals to watch include adoption of SymFold or similar architectures in protein design pipelines, further benchmarks on diverse protein families, and potential integration with experimental validation studies. The paper suggests a promising direction for the community, likely prompting follow-up research on symmetric multimodal architectures.