Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..
FIGRDock: Fast Interaction-Guided Regression for Flexible Docking
Authors: Shikun Feng, Bicheng Lin, Yuanhuan Mo, Yuyan Ni, Wenyu Zhu, Bowen Gao, Wei-Ying Ma, haitao li, Yanyan Lan
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experimental results show that FIGRDock achieves up to 100 faster inference than diffusion-based docking methods, while consistently surpassing them in accuracy across standard benchmarks. These results suggest that FIGRDock has the potential to offer a scalable and efficient solution for flexible docking, advancing the pace of structure-based drug discovery. |
| Researcher Affiliation | Academia | 1Institute for AI Industry Research (AIR), Tsinghua University, Beijing, China 2Zhongguancun Institute of Artificial Intelligence, China 3School of Basic Medical Sciences, Tsinghua University, China 4School of Software Engineering, South China University of Technology 5Academy of Mathematics and Systems Science, Chinese Academy of Sciences 6Beijing Frontier Research Center for Biological Structure, Tsinghua University, Beijing, China 7 Beijing Academy of Artificial Intelligence, Beijing, China |
| Pseudocode | No | The paper describes the methodology with figures and text but does not include a distinct section or block labeled as 'Pseudocode' or 'Algorithm'. |
| Open Source Code | Yes | The code is open-sourced in link https://github.com/fengshikun/FIGRDock.git |
| Open Datasets | Yes | We leverage the SIU dataset [7], which contains a substantial amount of synthetic computational complex data generated by docking software, as pre-training data to learn interaction-informed paired representations between the protein and ligand. ... In the fine-tuning stage, we fine-tune FIGRDock on the commonly adopted PDBbind v2020 dataset[22], which contains 19K crystal complex structures. ... We evaluate FIGRDock on the PDBbind test set and the Pose Busters V2 [3] test set. |
| Dataset Splits | Yes | We employ the time-split of PDBbind with 17k complexes from 2018 or earlier for training and validation, and 363 test structures from 2019, ensuring consistency with previous works[19, 4]. |
| Hardware Specification | Yes | The pre-training was conducted on 4 GPUs for 10 epochs with a batch size of 64, taking approximately 6 days to complete. ... The training was conducted using 4 A100 GPUs with a batch size of 16, and the pre-training took approximately 20 days to complete. ... The fine-tuning is performed on 4 A100 GPUs for 100 epochs with a batch size of 16, taking approximately 3 days to complete. |
| Software Dependencies | No | The input apo ligand conformation is generated using RDKit with a random seed, while the input apo protein structure is predicted by ESMFold [10]. The paper mentions software tools like RDKit, ESMFold, Uni-Mol, but does not specify their version numbers. |
| Experiment Setup | Yes | Table 4: Hyperparameter settings for Pocket Pretraining, Conditional Pretraining, and Fine-tuning stages. Batch Size 64 16 16 Training Epochs 10 12 100 Learning Rate 1 10 4 3 10 4 3 10 4 LR Scheduler polynomial_decay polynomial_decay polynomial_decay Warmup Ratio 0.01 0.06 0.06 Optimizer Adam Adam Adam Weight Decay 1 10 4 0 0 GPU Number 4 4 4 |