# MALDI-TOF MS in conjunction with machine learning: toward a new era for antimicrobial susceptibility testing

## TL;DR

This paper explores how MALDI-TOF MS combined with machine learning can speed up antimicrobial resistance testing.

## Contribution

The paper introduces a novel integration of MALDI-TOF MS and machine learning for rapid antimicrobial resistance prediction.

## Key findings

- Large-scale datasets can comprehensively reflect true resistance status.
- Stacking ensemble learning methods improve predictive performance.
- Model interpretation enhances transparency and acceptability.

## Abstract

Global public health is formidably threatened by antimicrobial resistance (AMR). Antimicrobial susceptibility testing (AST) is characterized by its long duration. Matrix-assisted laser desorption/ionization time-of-flight mass spectrometry (MALDI-TOF MS) is notable for its rapid analysis and cost-effectiveness. However, its role in AST has not been fully explored. In recent years, new opportunities for predicting AMR using MALDI-TOF MS data have been provided by the development of machine learning (ML) technologies. The research progress in using MALDI-TOF MS combined with ML for AMR testing is surveyed by this review, and critical steps including raw MALDI-TOF MS data acquisition, raw data preprocessing, algorithm selection, hyperparameter optimization, among others. It was found by us that the true resistance status can be comprehensively reflected by large-scale datasets, but effective management of high-dimensional data challenges is required. Algorithm performance can be enhanced by identifying the optimal combination of hyperparameters. Better predictive performance than individual models can be achieved by stacking ensemble learning methods. Model performance and generalizability can be more effectively assessed by metrics such as the Area Under the Receiver Operating Characteristic Curve (AUROC). The decision-making process can be understood by users with the help of model interpretation, thereby increasing model transparency and acceptability. Insufficient sample size, inadequate data standardization, and limited model generalizability are included in the current challenges. Continuously optimized, the integration of MALDI-TOF MS and ML is poised to open future avenues for rapid and accurate AMR prediction.