Adapts Casanovo for de novo sequencing with prediction of specific Post-Translational Modifications (PTMs).
Note: Identifies only known, predefined PTMs listed in config.yaml. Cannot perform open modification searches.
- Predicts PTMs defined in
config.yamlusing a fixed vocabulary (each AA+PTM is a unique token). - Includes input annotation standardization (
standardize_sequence.py) for robust training. - Generates detailed mzTab output with PTM locations, scores, and total mass shifts.
- Built upon the Casanovo (
depthcharge) Transformer architecture.
PTMs are handled by defining all target AA+PTM combinations (e.g., C+57.02146) with their masses in the config.yaml residues section. This creates a fixed model vocabulary. The standardize_sequence.py utility maps input annotations to these exact tokens for training. The model predicts sequences using these tokens. The MztabWriter then parses these predicted tokens to generate mzTab PTM information, including calculated total mass shifts.
- No Open Modification Search: Only identifies PTMs explicitly defined in
config.yaml. - Vocabulary Size: Large numbers of defined PTMs increase vocabulary size and potentially training difficulty.
- Clone the repository:
git clone <your-repository-url> cd <repository-directory>
- Install dependencies:
(Assuming you have a requirements.txt file)
pip install -r requirements.txt
- Change the run.sh script for desired function:
bash run.sh
- Modify the
residuessection to include all standard amino acids and the specific AA+PTM combinations you want the model to be able to predict. Ensure the masses are accurate. - You can copy the default configuration and modify it:
cp src/config.yaml my_custom_config.yaml # Now edit my_custom_config.yaml - Specify your custom config file using the
-cor--configoption when running commands.
The main entry point is src/denovo_ptm/denovo.py.
# Show help message and commands
python -m src.denovo_ptm.denovo --help