Proteus: Exploring Protein Structure Generation for Enhanced Designability and Efficiency
Authors: Chentong Wang, Yannan Qu, Zhangzhi Peng, Yukai Wang, Hongli Zhu, Dachuan Chen, Longxing Cao
ICML 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We have validated our model s performance on de novo protein backbone generation through comprehensive in silico evaluations and experimental characterizations, which demonstrate a remarkable success rate. |
| Researcher Affiliation | Collaboration | 1Zhejiang University, Hangzhou, Zhejiang, China 2School of Life Sciences, Westlake University, Hangzhou, China 3Duke University, Durham, North Carolina, USA. |
| Pseudocode | Yes | Algorithm 1 Proteus Model Inference |
| Open Source Code | Yes | Codes are available at https:// github.com/Wangchentong/Proteus. |
| Open Datasets | Yes | We curated a dataset from the Protein Data Bank (PDB) (Berman et al., 2000) with a cutoff date of August 1, 2023. |
| Dataset Splits | No | The paper mentions training data and evaluation, but does not provide specific details on how the dataset was split into training, validation, and test sets (e.g., percentages, sample counts, or citations to predefined splits). |
| Hardware Specification | Yes | Efficiency is estimated as the time taken to generate a protein backbone on a standard NVIDIA Ampere Tesla A40 GPU with 48 GB of GPU memory. |
| Software Dependencies | No | The paper mentions using optimizers like Adam and tools like Protein MPNN and ESMFold, citing their respective papers. However, it does not provide specific version numbers for any software libraries or packages (e.g., PyTorch 1.x, TensorFlow 2.x, numpy 1.x) used for implementation. |
| Experiment Setup | Yes | Table 6: Model Training Parameters provides specific details: 'Dimension of sequence track Cs 256', 'Max batch size 16', 'Training steps 2*10^6', 'Learning rate 0.0001', 'Optimizer Adam(β1=0.9, β2=0.999)'. |