Abstract
Imaging mass spectrometry enables untargeted spatial profiling of compounds in animal tissues. Data preprocessing and mining are central to comprehensively unravel the complexity of hyperspectral imaging mass spectrometry experiments. Herein, we describe a user-friendly, partially GPU- or compiler-accelerated software pipeline that enables multi-ROI, multi-condition, and multi-replicate preprocessing and mining of larger-than-memory imaging mass spectrometry datasets in Python. The package, termed iMSminer, streamlines computational imaging mass spectrometry workflows, from spectral preprocessing to unsupervised exploratory analysis to univariate fold-change statistical analysis. These capabilities enable mining of ions for molecular co-localization, characteristic molecular profiles, and differential expression. Functions include raw imzML import, peak picking, baseline subtraction, mass alignment, peak integration, normalization, ROI selection, calibration, chemical database search, analyte filtering, image processing, box plot visualization, volcano plot and heatmap visualizations, dimensionality reduction, image clustering, and in situ segmentation. Furthermore, data processed by iMSminer can be easily interfaced to standard deep learning packages and other special-purpose modelling tasks in Python for more advanced use cases.
Funding Information
  • National Institute of General Medical Sciences (R01 GM138660)
  • University of Florida (CLAS Scholars Award)
  • Bristol-Myers Squibb (Science Scholars Award)
  • University of Florida (Keaffaber Scholar Award)
  • Barry Goldwater Scholarship and Excellence in Education Foundation (Barry Goldwater Scholarship)

This publication has 18 references indexed in Scilit: