SurfDesign: Effective Protein Design on Molecular Surfaces
Keywords: Molecular Surfaces, Protein Design, Geometric Deep Learning
Abstract: Structure-based inverse folding has been extensively explored in recent years. In contrast, surface-conditioned protein generation is still an under-explored area. Molecular surfaces characterized by a compact and smooth composition of atoms at their boundary hold a more direct relevance to biomolecular interactions and function.
In this work, we introduce a novel framework named SurfDesign with several key improvements. Firstly, considering the theoretical fact that the molecular surface is a continuous manifold with infinite resolution, we propose surface-based equivariant message passing (SEMP) to incorporate the normal vector and curvatures and get aware of the manifold's Euclidean locality. Besides, a hybrid parameter-efficient fine-tuning (PEFT) technique is employed to combine the knowledge of protein language models (PLMs) with the surface geometric encoder. We extensively evaluate SurfDesign on the CATH, TS50, TS500, and PDB datasets, achieving an average recovery of more than 70\%. Our work opens another road to designing functional proteins, underscoring the importance of including surface attributes in conventional inverse folding.
Primary Area: applications to physical sciences (physics, chemistry, biology, etc.)
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 1826
Loading