Fast and Accurate Blind Flexible Docking

Authors: Zizhuo Zhang, Lijun Wu, Kaiyuan Gao, Jiangchao Yao, Tao Qin, Bo Han

ICLR 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments on public benchmark datasets demonstrate that FABFlex not only achieves superior effectiveness in predicting accurate binding modes but also exhibits a significant speed advantage (208 ) compared to existing state-of-the-art methods. Our experiments are conducted on public docking benchmark PDBBind to evaluate our FABFlex against a variety of docking methods. Table 1: Ligand performance comparison of blind flexible docking.
Researcher Affiliation Collaboration Zizhuo Zhang1 Lijun Wu2 Kaiyuan Gao3 Jiangchao Yao4 Tao Qin5 Bo Han1 1TMLR Group, Department of Computer Science, Hong Kong Baptist University 2Shanghai AI Laboratory 3Huazhong University of Science and Technology 4CMIC, Shanghai Jiao Tong University 5Microsoft Research AI4Science
Pseudocode Yes The pseudo code is provided in Appendix A.2. Algorithm 1 Pseudo code of FABFlex s inference.
Open Source Code Yes Our code is released at https://github.com/tmlr-group/FABFlex.
Open Datasets Yes Our experiments are conducted on the widely used public PDBBind v2020 dataset1, which contains a comprehensive collection of 19,443 protein-ligand crystal complex structures with experimentally measured binding affinities. 1http://pdbbind.org.cn/
Dataset Splits Yes Specifically, complexes deposited before 2019 are utilized as the training set (12,807 complexes) and validation set (734 complexes), while those recorded after 2019 are designated as test set (303 complexes).
Hardware Specification Yes The model is trained on eight NVIDIA RTX 4090 GPUs.
Software Dependencies No The experiments are conducted using the Pytorch framework5. The dimension dl of initial features extracted via Torch Drug (Zhu et al., 2022) for ligand atoms is set to 56, and the dimension dp of ESM-2 (Lin et al., 2022) features for amino acid is 1280. While PyTorch is mentioned, specific version numbers for PyTorch, Torch Drug, or ESM-2 are not provided.
Experiment Setup Yes Table 5: Implementation configuration of FABFlex. The more details of hyperparameter setting are provided in Appendix B.3.