A systematic review of machine learning on clinical MALDI-TOF MS

Area of Science:

• Clinical microbiology
• Bioinformatics
• Machine learning applications

Background:

• Bacterial identification, antimicrobial resistance prediction, and strain typification are crucial in clinical microbiology.
• Machine learning (ML) shows potential to improve Matrix-Assisted Laser Desorption/Ionization-Time of Flight Mass Spectrometry (MALDI-TOF MS) for these tasks.
• A comprehensive technical ML review for MALDI-TOF MS applications is lacking.

Purpose of the Study:

• To systematically review ML applications in MALDI-TOF MS for bacterial diagnostics.
• To identify key ML aspects including data, preprocessing, models, and availability.
• To provide insights and guidelines for enhancing ML-driven bacterial diagnostics.

Main Methods:

• Systematic literature review of 115 studies (2004-2025).
• Focused on ML aspects: data size/balance, preprocessing, model selection/evaluation, data/code availability.
• Analysis of classical ML and deep learning approaches.

Main Results:

• Classical ML models (Random Forest, SVM) are predominant; deep learning is emerging.
• Challenges include inconsistent preprocessing, black-box models, limited external validation, and poor open-source resource availability.
• These issues impede transparency, reproducibility, and adoption.

Conclusions:

• Standardized methodologies, increased transparency, and better data accessibility are needed.
• Actionable insights are provided to improve ML in MALDI-TOF MS analysis.
• Guidelines are offered for researchers on ML model development and evaluation for bacterial diagnostics.