Deep Reinforcement Learning for Modelling Protein Complexes
Authors: Ziqi Gao, Tao Feng, Jiaxuan You, Chenyi Zi, Yan Zhou, Chen Zhang, Jia Li
ICLR 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | 4 EXPERIMENTS Quantitative results show that GAPN achieves superior structure prediction accuracy compared to advanced baseline methods, with a speed-up of 600 . Moreover, we contribute a dataset containing non-redundant complexes. |
| Researcher Affiliation | Collaboration | Ziqi Gao HKUST Tao Feng HKUST (GZ) Jiaxuan You UIUC Chenyi Zi HKUST (GZ) Yan Zhou Createlink Technology Chen Zhang Createlink Technology Jia Li HKUST (GZ) |
| Pseudocode | Yes | Algorithm 1 GAPN Input: ground truth assembly graphs Gr. |
| Open Source Code | No | The paper does not provide an explicit statement or a link for open-source code for the methodology described. |
| Open Datasets | Yes | Following Mo LPC, we conduct experiments on protein complexes with the chain number 3 N 30. We download all available complexes of their first biological assembly version from the PDB database (Berman et al., 2000). |
| Dataset Splits | Yes | Finally, we obtained 7,063 PDBs for the training set and validation set, and 180 for the test set. Table 5: The dataset statistic after complete data processing and filtering. Chain number Train Valid Test ... Overall 6054 1009 180 |
| Hardware Specification | Yes | We test all baselines using 2 GeForce RTX 4090 GPUs. |
| Software Dependencies | No | The paper mentions software components like ESM and Adam Optimizer but does not provide specific version numbers for these or other software dependencies like Python, PyTorch, or CUDA. |
| Experiment Setup | Yes | Table 7: Hyperparameter choices of GAPN. It lists specific values for GCN layer number (2), MLP dimensions (e.g., 13, 32,32), Dropout rate (0.2), Clip ratio (0.2), Batch size (100), and Initial learning rates (e.g., 3e-4) for various components. |