maldi-nn

Deep learning tools and models for MALDI-TOF mass spectra analysis.

Package features:

• Reading and preprocessing functions for MALDI-TOF MS spectra.
• Model definitions to process SMILES strings with state-of-the-art techniques (for feature-based AMR prediction).
• Model definitions to pre-train state-of-the-art Transformer networks on MALDI-TOF MS data
• Model definitions and scripts to train AMR models on the DRIAMS database.
• Model definitions and scripts to train species identification models.

Install

maldi-nn is distributed on PyPI.

pip install maldi-nn

In case this package loses backward-compatibility with more-recent versions of PyTorch and PyTorch Lightning: the code has been tested with torch==2.0.1 and pytorch-lightning==2.0.9. If you encounter errors with these packages, try running the code using these versions.

You may need to install PyTorch before running this command in order to ensure the right CUDA kernels for your system are installed

Quick start

We have uploaded some example spectra from the RKI database and our pre-trained models in the assets folder.

To quickly start playing around with our models, follow:

In bash:

pip install maldi-nn
git clone https://github.com/gdewael/maldi-nn.git

In Python:

from maldi_nn.spectrum import *
from maldi_nn.models import MaldiTransformer
import torch
spectrum = SpectrumObject.from_bruker(
    "./maldi-nn/assets/RKI_example/Bacillus_anthracis/acqu",
    "./maldi-nn/assets/RKI_example/Bacillus_anthracis/fid"
    )

preprocessor = SequentialPreprocessor(
    VarStabilizer(method="sqrt"),
    Smoother(halfwindow=10),
    BaselineCorrecter(method="SNIP", snip_n_iter=20),
    Trimmer(),
    PersistenceTransformer(extract_nonzero=True),
    Normalizer(sum=1),
    PeakFilter(max_number=200),
)

spectrum_preprocessed = preprocessor(spectrum)
spectrum_tensors = spectrum_preprocessed.torch()

model = MaldiTransformer.load_from_checkpoint("../../maldi-nn/assets/MaldiTransformerM.ckpt").eval().cpu()

mlm_logits, spectrum_embedding = model(spectrum_tensors)

prob_noise_peak = torch.sigmoid(mlm_logits)

Other Maldi Transformer model sizes are available at https://huggingface.co/gdewael/MaldiTransformer/tree/main.

Academic Reproducibility

This package contains all code and scripts to reproduce: "An antimicrobial drug recommender system using MALDI-TOF MS and dual-branch neural networks", and "Pre-trained Maldi Transformers improve MALDI-TOF MS-based prediction". All information regarding reproducing our results can be found in the reproduce folder README

Credits

• Implementations of many MALDI reading and processing functions were based on the R package MaldiQuant.
• Topological Peak Filtering was taken from the Topf package.

Citation

Antimicrobial drug recommenders:

@article{dewaele2023antimicrobial,
  title={An antimicrobial drug recommender system using MALDI-TOF MS and dual-branch neural networks},
  url={http://dx.doi.org/10.1101/2023.09.28.559916},
  DOI={10.1101/2023.09.28.559916},
  publisher={Cold Spring Harbor Laboratory},
  author={De Waele, Gaetan and Menschaert, Gerben and Waegeman, Willem},
  year={2023},
  month=sep,
}

Maldi Transformers:

@article{dewaele2024pre,
  title={Pre-trained Maldi Transformers improve MALDI-TOF MS-based prediction},
  author={De Waele, Gaetan and Menschaert, Gerben and Vandamme, Peter and Waegeman, Willem},
  journal={bioRxiv},
  pages={2024--01},
  year={2024},
  publisher={Cold Spring Harbor Laboratory}
}