PNNL-m-q / qcgnoms

A Graph Neural Net for ms/ms prediction.

QC-GN²OMS² - A Graph Neural Net for ms/ms prediction.

Prediction:

• Generate graph feature database in a pickle file
  • top_db.py : creates TOP feature database using RDKit
  • qc1_3d.py : generates 3D molecular structures for the QC1 database
  • qc1_db.py : aggregates the 3D structures and calculates features
  • qc2_db.py : calculates BDE features for the QC2 model and stores them in a pickle file
• Run the prediction script with the associated model and database
  • pred_afb.py : predict a spectra using the QC2 model
• Example:

cd weights/
cat qc2_1.model.a* > qc2_1.model
cd ../example
python ../qc2_db.py example.csv ex.pkl
python ../pred_qc2.py ./ex.pkl 30
cat pred_qc2.ms

Training:

Requires a MS/MS database in a pickle file. Required data columns:

1. Collision Energy in eV
2. InChI
3. Smiles
4. M/Z: numpy array of high resolution m/z values.
5. Intensity: numpy array of MS intensities.

• See data/msms_sample.pkl

• Generate graph feature database in a pickle file

  • top_db.py : creates TOP feature database using RDKit
  • qc1_3d.py : generates 3D molecular structures for the QC1 database
  • qc1_db.py : aggregates the 3D structures and calculates features
  • qc2_db.py : calculates BDE features for the QC2 model and stores them in a pickle file

• Test datasets are assigned by first training the control model with train_control.py

• Test set data are located in test_set/

Dependencies

• torch
• torch-geometric
• alfabet https://github.com/NREL/alfabet
• xtb https://github.com/grimme-lab/xtb
• openbabel
• pandas
• matplotlib
• numpy

Citation

QC-GN²oMS²: a Graph Neural Net for High Resolution Mass Spectra Prediction
Richard Overstreet, Ethan King, Julia Nguyen, Danielle Ciesielski bioRxiv 2023.01.16.524269; doi: https://doi.org/10.1101/2023.01.16.524269