Zero-Shot 3D Drug Design by Sketching and Generating
Authors: Siyu Long, Yi Zhou, Xinyu Dai, Hao Zhou
NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experiments show that DESERT achieves a new state-of-the-art at a fast speed.1 |
| Researcher Affiliation | Collaboration | Siyu Long 1, Yi Zhou2, Xinyu Dai1, Hao Zhou3 1National Key Laboratory for Novel Software Technology, Nanjing University 2Byte Dance AI Lab 3Institute for AI Industry Research (AIR), Tsinghua University |
| Pseudocode | Yes | We present an algorithm (see the Appendix 1.1) to obtain the desired molecule shape, which is of the appropriate size and complementary to the pocket surface. |
| Open Source Code | Yes | Code is available at https://github.com/longlongman/DESERT. |
| Open Datasets | Yes | We use 100M unbound molecules sampled from the lead-like subset of ZINC as the training data. |
| Dataset Splits | No | The paper specifies training data and test data, but does not provide specific details on a separate validation set split (e.g., percentages or counts) used during training or hyperparameter tuning. |
| Hardware Specification | Yes | The model is developed by Para Gen [68] 6 and trained on 32 Telsa V100 GPU cards for 2 weeks. |
| Software Dependencies | No | The paper mentions several software tools and libraries (e.g., Adam W, Para Gen, RDKit, Vina), but it does not provide specific version numbers for these software dependencies, which would be necessary for full reproducibility. |
| Experiment Setup | Yes | When training SHAPE2MOL, we set the dropout rate as 0.1, batch size 2048, train step 300K and use Adam W [67] with learning rate 5e-4, weight decay 1e-2, and warmup step 4000 as the optimizer. |