Sovereign 8s.is mirror of standardgalactic/Pred-MutPRI
- Python 94.3%
- Shell 5.7%
| Filename | Latest commit message | Latest commit date |
|---|---|---|
| alphafold3 | ||
| Cytoscape | ||
| Data | ||
| dssp | ||
| FoldX | ||
| ProtInter | ||
| Scripts | ||
| x3dna-dssr | ||
| CITATION.cff | ||
| LICENSE | ||
| model.json | ||
| README.md | ||
External dependencies (download links)
This pipeline relies on several external tools. Please install/download them first and make sure they are available as described below.
FoldX
- Website: https://foldxsuite.crg.eu/
- Download/License: https://foldxsuite.crg.eu/academic-license-info
DSSP / mkdssp
- DSSP (mkdssp) source and documentation: https://github.com/PDB-REDO/dssp
- Note: On some systems the executable may be named
mkdsspinstead ofdssp.
ProtInter
- Repository (ProtInter): https://github.com/maxibor/protinter
DSSR (x3dna-dssr)
- 3DNA/DSSR homepage: http://x3dna.org/
Cytoscape (for network features)
- Cytoscape homepage: https://cytoscape.org/
- Download page: https://cytoscape.org/download/
- Important: Cytoscape must be running (opened) during prediction, so that the pipeline can connect to it and compute network features.
AlphaFold3 (for ES feature)
- AlphaFold3 (official): https://github.com/google-deepmind/alphafold3
- Note: Please follow the official installation and model-parameter download instructions.
Environments
AlphaFold3 dependencies may conflict with ESM (and other Python packages used for prediction).
Therefore, we created two separate conda environments:
af3: used to ensure AlphaFold3 runs correctly (ES feature).env1: the main prediction execution environment (feature extraction + XGBoost prediction, including ESM-related steps).
Python dependencies
Install the required Python packages in the env1 environment (example using pip):
pip install numpy pandas scikit-learn xgboost biopython
pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118
pip install fair-esm
pip install py4cytoscape networkx
Notes:
- The exact PyTorch command depends on your CUDA version (or use CPU-only wheels).
py4cytoscaperequires Cytoscape to be running for the network feature step.
Run prediction
From the project root directory, run:
./predict.sh -PDB 1A1T -CHAIN A -Mut E21K --model_json model.json
Replace the arguments with your own inputs:
-PDB: PDB ID (e.g.,1A1T)-CHAIN: chain ID (e.g.,A)-Mut: mutation in the formatWT<pos>MUT(e.g.,E21K)--model_json: path to the trained XGBoost model file
Notes / Tips
- The input PDB file is recommended to be a protein–RNA dimer complex structure (i.e., the mutated protein chain and the RNA chain are both present in the same PDB file) to reduce potential issues during feature extraction.
- Because the pipeline integrates AlphaFold3, prediction may take a noticeable amount of time.
For example, on a machine with 8 GB GPU memory (RTX 4060), a single prediction typically takes ~20 minutes on average. - Software environment: Please use the software versions and runtime environment specified in this repository whenever possible. Differences in operating systems or versions of external dependencies (e.g., FoldX, Cytoscape/py4cytoscape, DSSP, ProtInter, AlphaFold3, and related Python packages) may lead to slight variations in intermediate features and final predictions.