MS2MACCS predicts MACCS fingerprints (167 bit vector) from tandem mass spectrometry data (MS2). It does so by combining a look-up table for common fragment structure combined with a transformer architecture.

✨ Why use MS2MACCS?

➕➖ predicts fingerprints from spectra measured in positive and negative mode
⚡predicts ~40-300 spectra per second (cpu i7 8th generation vs gpu RTX A5000)
🔝 shows comparable results with other state-of-the art tools (we do not achieve as good results as Sirius)
🆓 MS2MACCS is completely open source (do with it whatever you want)

🛠 Installation

conda create -n ms2maccs python=3.10
conda activate ms2maccs

git clone https://github.com/j-a-dietrich/MS2MACCS.git
cd MS2MACCS
pip install -e .

# if cuda available
pip install torch==2.7.1+cu118 --index-url https://download.pytorch.org/whl/cu118

# if no cuda
pip install torch==2.7.1

🚀 Quickstart

from ms2maccs import MS2MACCS

m = MS2MACCS(
    "../models/standard_model.pt", 
    "../bit_maps/fp_bit_map_H_p_mode.pkl", 
    "../bit_maps/fp_bit_map_H_n_mode.pkl", 
    "cpu", 
)

pred_maccs = m.calc_fp("../ms2_data/test_specs_H_p_mode.mgf").to("cpu")

# for tox predictions are MLinvitroTox MACCS models necessary (no published yet)
pred_tox = m.calc_tox("../ms2_data/test_specs_H_p_mode.mgf").to("cpu")

# see demo/demo.ipynb

📬 Get in Touch

💡 Questions, ideas, or contributions? Open an issue.

📚 Citation

coming at some point (potentially)