Kernel density-encoded MALDI-TOF spectra with LightGBM machine learning for multiplex resistance prediction in Klebsiella pneumoniae.

Authors

Keqing Zhang, Heng Shen, Yanyan Hu, Zelin Yan, Linping Zhang, Rong Zhang

Published in

Diagnostic microbiology and infectious disease. Volume 116. Issue 4. Pages 117581. Jul 28, 2026. Epub Jul 28, 2026.

Abstract

Antimicrobial resistance(AMR) in Klebsiella pneumoniae is an urgent clinical challenge. Matrix-Assisted Laser Desorption/Ionization Time-of-Flight Mass Spectrometry(MALDI-TOF-MS) is routinely used for species identification, and reusing these spectra for same-day AMR prediction could accelerate targeted therapy. We developed and validated Light Gradient Boosting Machine(LightGBM) models to predict K. pneumoniae resistance directly from routine Bruker MALDI-TOF MS spectra of single bacterial colonies obtained on routine Bruker instruments, using a kernel density-encoded feature representation. Raw spectra were processed to extract peak-level attributes, including m/z, signal-to-noise ratio, peak area and intensity. The m/z axis was intervalized using a kernel density-guided strategy to preserve local spectral density and ordering, and peak attributes within each interval were aggregated into a structured high-dimensional matrix. A total of 424 isolates were included, and balanced binary datasets were constructed for 12 antibiotics, with an average of 98 ± 8 susceptible and 98 ± 8 resistant isolates per antibiotic; intermediate isolates were not included. The dataset included both environmental and human-derived isolates, and antimicrobial susceptibility labels were determined by broth microdilution for amikacin, aztreonam, ciprofloxacin, meropenem, piperacillin-tazobactam, cefepime, cefmetazole, cefoperazone-sulbactam, cefotaxime, ceftazidime, imipenem, and ceftazidime-avibactam, with stronger agreement for cefotaxime and comparatively lower agreement for imipenem. In internal hold-out validation, model accuracy ranged from 0.81 to 0.92 across 12 evaluable antibiotic-specific models, with AUROC values ranging from 0.88 to 0.96. These findings suggest that kernel density-encoded MALDI-TOF MS spectra combined with LightGBM may provide a proof-of-concept framework for AMR prediction, although independent external validation is required before clinical application can be considered.

PMID:
42551115
Bibliographic data and abstract were imported from PubMed on 05 Aug 2026.

Read full publication at:
Please sign in to see all details.

Stats

1-terrible, 9-excellent. How would you rate this publication? Sign in in to submit your rating.

Post a comment

You need to be signed in to post comments. You can sign in here.

Comments

There are no comments yet.