Early administration of effective antimicrobial treatments is critical for the outcome of infections and the prevention of treatment resistance. Antimicrobial resistance testing enables the selection of optimal antibiotic treatments, but current culture-based techniques can take up to 72 hours to generate results. We have developed a novel machine learning approach to predict antimicrobial resistance directly from MALDI-TOF mass spectra profiles of clinical samples. We trained calibrated classifiers on a newly-created publicly available database of mass spectra profiles from clinically most relevant isolates with linked antimicrobial susceptibility phenotypes. The dataset combines more than 300,000 mass spectra with more than 750,000 antimicrobial resistance phenotypes from four medical institutions. Validation against a panel of clinically important pathogens, including Staphylococcus aureus, Escherichia coli, and Klebsiella pneumoniae, resulting in AUROC values of 0.80, 0.74, and 0.74 respectively, demonstrated the potential of using machine learning to substantially accelerate antimicrobial resistance determination and change of clinical management. Furthermore, a retrospective clinical case study found that implementation of this approach would have resulted in a beneficial change in the clinical treatment in 88% (8/9) of cases. MALDI-TOF mass spectra based machine learning may thus be an important new tool for treatment optimization and antibiotic stewardship.

The DRIAMS dataset is ressource intended for antimicrobial resistance prediction from real-world clinical routine MALDI-TOF mass spectra. It is comprised of four subdatasets collected at different medical institutions across Switzerland. 

For each site, the data consists of MALDI-TOF mass spectra in the form of .txt files and a meta-data file.

(i) The meta-data, incl. species and antimicrobial resistance corresponding to each spectra, is part of the "id" folder

(ii) The remaining folders store the MALDI-TOF mass spectra in various stages of preprocessing: "raw" all spectra as extracted from the MALDI-TOF MS instrument, "preprocessed" all spectra after the application of an established preprocessing pipeline and "binned_6000" all spectra after the application of an established preprocessing pipeline and binning along the mass-to-charge-ratio axis with a bin size of 3Da, resulting in 6000 feature bins.

For details on the dataset extraction, quality control, preprocessing and properties, please refer to the Methods section in the corresponding publication at https://doi.org/10.1038/s41591-021-01619-9.

When using the data, please also cite the corresponding Nature Medicine article the following way:

Weis, C. et al. Direct antimicrobial resistance prediction from clinical MALDI-TOF mass spectra using machine learning. Nat Med (2022). https://doi.org/10.1038/s41591-021-01619-9

We recommend using our Python package for MALDI-TOF preprocessing and machine learning analysis, maldi-learn (https://github.com/BorgwardtLab/maldi-learn), to load and analyse DRIAMS data.

The github package comes with an elaborate README file, which gives details on installation and usage examples. In order to use this package the locations of data files and folder structure must be preserved. Please note that all four downloaded data packages should be kept in one folder, serving as the DRIAMS root folder, which then needs to be set as the DRIAMS_ROOT path in the .env file.

The folder structure obtained after download is the following:

DRIAMS
├── DRIAMS-A
│   ├── binned_6000
│   ├── id
│   ├── preprocessed
│   └── raw
├── DRIAMS-B
│   ├── binned_6000
│   ├── id
│   ├── preprocessed
│   └── raw
├── DRIAMS-C
│   ├── binned_6000
│   ├── id
│   ├── preprocessed
│   └── raw
└── DRIAMS-D
    ├── binned_6000
    ├── id
    ├── preprocessed
    └── raw