Awesome

Pythia

Structure-based self-supervised learning enables ultrafast prediction of stability changes upon mutations

Prerequisites

pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118
pip3 install -r requirements.txt

Usage

To use the Pythia, you can run it from the command line with the following options:

Basic Usage

cd pythia
python masked_ddg_scan.py

By default, this will process files in the directory ../s669_AF_PDBs/ using cuda:0 (GPU 0) if available.

Command Line Options

--input_dir: Specifies the directory path containing the PDB files. Default is ../s669_AF_PDBs/.

Example:
```
python masked_ddg_scan.py --input_dir "/path/to/directory/"
```
--pdb_filename: If you want to process a single PDB file instead of a directory, specify its path with this option.

Example:
```
python masked_ddg_scan.py --pdb_filename "/path/to/file.pdb"
```
--check_plddt: Use this flag if you want to filter PDB files based on their pLDDT value. Files with a pLDDT value less than the specified cutoff (see below) will be ignored.

Example:
```
python masked_ddg_scan.py --check_plddt
```
--plddt_cutoff: Specifies the pLDDT cutoff value if --check_plddt is used. Default is 95.

Example:
```
python masked_ddg_scan.py --check_plddt --plddt_cutoff 90
```
--n_jobs: Indicates the number of parallel jobs to run. Default is 2.

Example:
```
python masked_ddg_scan.py --n_jobs 4
```
--device: Specifies the device to use for computation. By default, it will use cuda:0 (GPU 0). If you want to use CPU or another GPU, specify it here. Valid values include cuda:0, cuda:1, ... for GPUs, or cpu for the CPU.

Example:
```
python masked_ddg_scan.py --device cpu
```

Examples

Process all PDB files in the directory /path/to/directory/, using the first GPU and checking pLDDT values with a cutoff of 90:
```
python masked_ddg_scan.py --input_dir "/path/to/directory/" --check_plddt --plddt_cutoff 90 --device cuda:0
```

Process a single PDB file /path/to/file.pdb using the CPU:

python masked_ddg_scan.py --pdb_filename "/path/to/file.pdb" --device cpu

Megascale dataset, S2648, S669 contains predictions and labels.

Train

Download preprocessed files for training at CATH dataset or BioA dataset from the Google Drive:
```
sbatch train_model.sh
```