Protein Conformation Generation via Force-Guided SE(3) Diffusion Models

Authors: Yan Wang, Lihao Wang, Yuning Shen, Yiqun Wang, Huizhuo Yuan, Yue Wu, Quanquan Gu

ICML 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments on a variety of protein conformation prediction tasks, including 12 fast-folding proteins and the Bovine Pancreatic Trypsin Inhibitor (BPTI), demonstrate that our method surpasses the state-of-the-art method.
Researcher Affiliation Collaboration 1Byte Dance Research 2School of Mathematical Sciences, Tongji University, Shanghai (this work was done during Yan s internship at Byte Dance Research) 3Department of Computer Science, University of California, Los Angeles (this work was done during Huizhuo and Yue s internship at Byte Dance Research). Correspondence to: Quanquan Gu <quanquan.gu@bytedance.com>.
Pseudocode Yes Algorithm 1 Force-guided CONFDIFF
Open Source Code No The paper does not contain an explicit statement or link indicating that the source code for the described methodology is publicly available.
Open Datasets Yes Training datasets. We train CONFDIFF using all available single-chain protein structures deposited to the PDB on or before Dec 31, 2021... fast-folding proteins comprise MD simulation data for 12 small proteins with fast folding-unfolding dynamics (Lindorff Larsen et al., 2021)... bovine pancreatic trypsin inhibitor (BPTI) contains MD simulation data for BPTI exhibiting five metastable states (Shaw et al., 2010).
Dataset Splits Yes We train CONFDIFF using all available single-chain protein structures deposited to the PDB on or before Dec 31, 2021, and perform validation using structures deposited between Jan 1, 2022 and Dec 31, 2022.
Hardware Specification Yes We used a single NVIDIA-V100 GPU to benchmark model performance.
Software Dependencies No The paper mentions several software tools like pyemma, Deeptime, Open MM, and faspr, along with citations, but does not specify their version numbers (e.g., 'pyemma 2.x' or 'Open MM 7.x').
Experiment Setup Yes Table S3. Hyperparameter choices of CONFDIFF... Batch Size 32 Learning Rate 1 10 4... During minimization, we keep the conformations by applying independent harmonic restraints on all heavy (non-hydrogen) atoms with spring constant of 10 kcal/mol A 2, the tolerance is set to 2.39 kcal/mol A 2 without maximal step limits.