MoDiff: a Protein Diffusion Model for Multi-motif Scaffolding Problem

Xiaoran Jiao, Weian Mao, Zheng Sun, Yu Bo, Peiyuan Yang, Hao Chen, Chunhua Shen

Keywords: Motif Scaffolding, Drug Design, Protein Modeling, Protein Diffusion Model

TL;DR: MoDiff, a new protein diffusion model, automatically designs multi-motif scaffolds for drug and enzyme design, overcoming inpainting limitations and enabling diverse scaffold generation.

Abstract:

The motif scaffolding problem, crucial in drug and enzyme design, involves creating a scaffold to support one or more specified motifs. Existing methods employ the paradigm of inpainting, thereby limiting their applicability to scenarios where only a single motif is present or the positions between multiple motifs are known. However, in many significant scenarios, the positions between motifs are unknown, leaving the multi-motif scaffolding problem open. To tackle this challenge, we introduce a protein diffusion model called MoDiff. During the diffusion process, MoDiff implicitly assigns motifs to the protein backbone, thereby achieving the automatic design of relative positions among motifs. Our experiments demonstrate that MoDiff can: 1) solve the multi-motif scaffolding problem even when the positions between motifs are unknown, and 2) generate diverse scaffolds based on multiple given motifs. This indicates that MoDiff is a potential general solution to the multi-motif scaffolding problem.

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: 7001