Modanovo is a de novo peptide sequencing tool for post-translationally modified (PTM) peptides, built on top of Casanovo (v4.0.0).

We recommend using a fresh conda environment with Python 3.10.

conda create --name modanovo-env python=3.10
conda activate modanovo-env

Install the dependencies:

1. PyTorch (pick the command matching your CUDA/CPU setup from the PyTorch site; generic example):

pip3 install torch

2. Depthcharge-MS (pinned commit):

pip install git+https://github.com/wfondrie/depthcharge.git@bd2861f

3. Clone this repository:

git clone https://github.com/gagneurlab/Modanovo.git
cd modanovo

4. Install Modanovo:

pip install .

For development (editable install):

pip install -e .

Modanovo supports three modes: training, evaluation, and inference.

Train a new model from scratch:

modanovo train -c <config_path> -p <val_paths> <train_paths>

Fine-tune from pretrained Casanovo weights:

modanovo train -c <config_path> -m <model_path> -p <val_paths> <train_paths>

Where <model_path> points to the pretrained Casanovo v4.0.0 weights.

Evaluate a trained model on validation/test spectra:

modanovo evaluate -c <config_path> -m <model_path> -p <val_paths>

Run Modanovo in inference mode:

modanovo sequence -c <config_path> -m <model_path> -o <out_path>

This writes peptide sequence predictions in .mzTab format.

Assuming you’ve installed Modanovo and have a model checkpoint:

# from the repo root
modanovo sequence \
  -c modanovo/config.yaml \
  -m path/to/casanovo_or_modanovo_weights.ckpt \
  -o outputs/predictions.mztab

Make sure that the defined residues are compatible with the model weights. Leaving the config entry expanded_residues in the configuration file empty uses Casanovo's tokens. By default, fine-tuning residues are those from the MULTI-PTM dataset in PROSPECT-PTM.

• Default configuration: modanovo/config.yaml
• Example spectra file: data_utils/example_data.mgf
• Train/val/test splits used during development: https://huggingface.co/datasets/gagneurlab/Modanovo-development-dataset
• Model weights for the model fine-tuned to cover 19 amino acid-PTM combinations: https://huggingface.co/gagneurlab/Modanovo-model

• Compatible with Casanovo v4.0.0 weights and formats.

• Casanovo: _Yilmaz, Melih, William E Fondrie, Wout Bittremieux, et al. 2024. “Sequence-to-Sequence Translation from Mass Spectra to Peptides with a Transformer Model.” Nature Communications 15 (1): 6427.

• PROSPECT-PTM: Gabriel, Wassim, Omar Shouman, Ayla Schroeder, Florian Boessl, and Mathias Wilhelm. 2024. “PROSPECT PTMs: Rich Labeled Tandem Mass Spectrometry Dataset of Modified Peptides for Machine Learning in Proteomics.” Advances in Neural Information Processing Systems 37.

If you use Modanovo in your research, please cite:

FIXME