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..
DynaPhArM: Adaptive and Physics-Constrained Modeling for Target-Drug Complexes with Drug-Specific Adaptations
Authors: Diya Zhang, Mengwei Sun, Xingdan Wang, Cheng Liang, Qiaozhen Meng, Shiqiang Ma, Fei Guo
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experimental results demonstrate that Dyna Ph Ar M achieves state-of-the-art performance with an overall root mean square deviation (RMSD) of 2.01 Å and a sc-RMSD of 0.29 Å while exhibiting higher success rates compared to existing methodologies. |
| Researcher Affiliation | Academia | 1Central South University 2Shandong Normal University 3Xiangtan University 4Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences |
| Pseudocode | Yes | Appendix D Cooperative scalar-vector representation algorithm Algorithm 2 Scalar-Vector Feature Interaction with SE(3)-Equivariant Update |
| Open Source Code | Yes | We provide anonymized code, datasets and detailed instructions in the anonymous link https://anonymous.4open.science/r/Dyna Ph Ar M-8352 to reproduce the main experimental results. |
| Open Datasets | Yes | We curated a comprehensive dataset of 21,762 high-quality target-drug complexes from the Protein Data Bank (PDB) and 2,417 validated drugs from Drug Bank. |
| Dataset Splits | Yes | To prevent data leakage and ensure generalizability, complexes were grouped by 30% sequence identity, with entire clusters assigned to either the training (80%) or validation (20%) sets, guaranteeing no sequence overlap between subsets. |
| Hardware Specification | Yes | Our proposed method, Dyna Ph Ar M, achieves the fastest average reconstruction time (2.39 s per complex on an NVIDIA RTX 4090) |
| Software Dependencies | No | The optimization of the model s parameters is performed using the Adam algorithm, as implemented in the Py Torch library. |
| Experiment Setup | Yes | The training process is initiated with a learning rate of 1 10 4, while the β1, β2, and ϵ hyperparameters of the Adam optimizer are maintained at their default Py Torch values of 0.9, 0.999, and 1 10 8, respectively. ... Table 8: Key hyperparameters used in the proposed model, covering embedding dimensions, interaction modules, diffusion and decoder configuration, physical constraints, and inference setup. LEARNING_RATE 1e-4 BATCH_SIZE 64 NUM_EPOCHS 50 |