Introduction

PathDiffusion is a software that automatically predicts protein folding pathways using evolution-guided diffusion models. It comprises two main modules: the first module focuses on preparing position-specific noise schedules (PSNS), and the second module employs PSNS-guided diffusion models to generate the folding pathway.

Model

sequence-conditional model

unconditional model

Training datasets

We train PathDiffusion using protein structures from the Protein Data Bank (for conditional generation) and the IDRome database (for unconditional generation). Details on dataset preparation can be found in https://github.com/YangLab-SDU/PathDiffusion

Benchmark datasets

FP52 dataset

MD12 dataset

LT20 dataset

IDP50 dataset

Reference

Zhao et al, PathDiffusion: modeling protein folding pathway using evolution-guided diffusion, bioRxiv 2026.

---