Zenodo description for DESI NMF deposit Data and code for: Automated separation of overlapping fingermarks by non-negative matrix factorization of DESI mass spectrometry imaging data Description This deposit contains the data and analysis code accompanying the manuscript: Villesen P, Nielsen K, Frisch K. "Automated separation of overlapping fingermarks by non-negative matrix factorization of DESI mass spectrometry imaging data. " Analytical and Bioanalytical Chemistry (2026). Background. Desorption electrospray ionization mass spectrometry imaging (DESI-MSI) can visualize the spatial distribution of hundreds to thousands of molecular species across latent fingermarks. When fingermarks from different donors overlap on the same surface, standard imaging approaches cannot separate the individual contributions. We apply non-negative matrix factorization (NMF) to decompose DESI-MSI datasets into spatially coherent components, each representing a distinct molecular signature. Components are ranked by Moran's I (a measure of spatial autocorrelation) so that the most spatially structured — and therefore most informative — components appear first. Data. Five gelatin-lifter slides containing latent fingermarks from 10 donors were imaged at 50 µm pixel resolution using a Waters Xevo G2-XS QToF mass spectrometer with a DESI XS ion source (mass range 50–1200 m/z, scan rate 200 µm/s). Each slide was acquired over 11–15 hours. The raw mass spectrometry imaging data were exported from Waters HDI 1. 6 as tab-delimited text files (top 5, 000 m/z features by total signal intensity, mass window 0. 02 Da). The five slides are: Slide 1 (230215): two overlapping fingermarks, donors D5/D6 Slide 2 (231019): two overlapping fingermarks, donors D6/D5 Slide 3 (231108): two overlapping fingermarks, donors D11/D12 Slide 4 (240202): two overlapping fingermarks, donors D1/D6 Slide 5 (240208): two overlapping fingermarks, donors D5/D1 A simulated four-fingermark overlay dataset was constructed by combining slides 1 and 3 (donors D5, D6, D11, D12) after cropping, 0. 1 Da m/z binning, and global intensity normalisation. Code. The analysis pipeline consists of four steps implemented in R Markdown: step₀1dataconversionᵥ12. rmd — Reads raw exported text files and converts them to compressed R data objects (. desi. qs) with log1p-transformed intensity matrices. step₀2createₚseudoₒverlayᵥ01. rmd — Constructs the simulated four-fingermark overlay dataset from two individual slides. step₀3PCAₐndNMFᵥ12. rmd — Performs PCA and NMF (via the RcppML package; alternating least squares, squared Frobenius norm loss, K = 30 components per slide, K = 40 for the overlay, 500 iterations, tolerance 1e-6). Components are ordered by Moran's I. step₀4ₚlottingₐndfiguresᵥ14. rmd — Generates all main and supplementary figures. A gwf workflow file (workflow. py) orchestrates the pipeline on an HPC cluster. An R initialisation script (pallesᵢnit. R) sets shared ggplot themes and utility functions. Software requirements. R (≥ 4. 0) with packages: RcppML, qs, spdep, tidyverse, ggplot2, patchwork, viridis, Matrix. Python 3 with gwf (for HPC workflow orchestration). Computations were performed on an HPC cluster (4 cores, 100 GB RAM per job; NMF runtime up to 25 minutes per slide). Related resources. Prior publication on the DESI-MSI fingermark method: Frisch K et al. (2024) Anal. Chem. 96: 12497–12506. DOI: 10. 1021/acs. analchem. 4c02626 Keywords DESI-MSI, mass spectrometry imaging, non-negative matrix factorization, NMF, fingermarks, fingerprints, forensic science, spatial autocorrelation, Moran's I, desorption electrospray ionization
Villesen et al. (Mon,) studied this question.