Scholarlink

Background <p>Carbapenem-resistant <i>Acinetobacter baumannii</i> (CRAB) is a major pathogen in hospital-acquired infections, and rapid detection is critical for guiding antimicrobial therapy and infection control. Conventional antimicrobial susceptibility testing (AST) has a long turnaround time, limiting early clinical decision-making.</p> Methods <p>We developed a rapid CRAB prediction model by integrating routine clinical matrix-assisted laser desorption/ionization time-of-flight mass spectrometry (MALDI-TOF MS) data with machine learning. A total of 301 <i>A. baumannii</i> isolates, including 189 CRAB and 112 carbapenem-susceptible <i>A. baumannii</i> (CSAB), generated 602 high-quality spectra. Train–test splitting was performed at the isolate level to prevent information leakage. Fixed binning and K-means clustering–based dynamic binning strategies were compared, with K-means-derived bin boundaries learned from the training set and fixed for independent test-set transformation. Logistic regression, support vector machine, random forest, Light Gradient Boosting Machine, and Extreme Gradient Boosting models were trained and optimized using feature selection uniformly performed on the entire training dataset prior to nested cross-validation for hyperparameter optimization, and their performance was evaluated on an independent test set. The optimal model was interpreted using SHapley Additive exPlanations (SHAP).</p> Results <p>No significant differences were observed between CRAB and CSAB in clinical sources or baseline characteristics. Dynamic binning improved predictive performance across models. Optimal performance was achieved with 650 bins and 50 retained features. Under this setting, Light Gradient Boosting Machine (LightGBM) achieved an area under the receiver operating characteristic curve (ROC-AUC) of 0.985 and an accuracy of 0.942 on the test set. SHAP analysis identified multiple m/z intervals and statistical features that contributed to CRAB discrimination.</p> Conclusions <p>This study presents a rapid CRAB detection strategy based on MALDI-TOF MS and machine learning. Using dynamic binning and optimized feature selection, the LightGBM model achieved excellent predictive performance, with a ROC-AUC of 0.985 and an accuracy of 0.942 on the independent test set. The framework requires no additional experimental procedures, provides rapid predictions, and offers enhanced interpretability through SHAP analysis. The proposed framework warrants further validation in clinical settings for rapid antimicrobial resistance screening.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Rapid identification of carbapenem-resistant Acinetobacter baumannii based on MALDI-TOF mass spectrometry and machine learning

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年6月17日

查看全文
  • Meng Zhou,
  • Liru Zhao,
  • Dongbo Li,
  • Zhongle Cheng¹

摘要

Background

Carbapenem-resistant Acinetobacter baumannii (CRAB) is a major pathogen in hospital-acquired infections, and rapid detection is critical for guiding antimicrobial therapy and infection control. Conventional antimicrobial susceptibility testing (AST) has a long turnaround time, limiting early clinical decision-making.

Methods

We developed a rapid CRAB prediction model by integrating routine clinical matrix-assisted laser desorption/ionization time-of-flight mass spectrometry (MALDI-TOF MS) data with machine learning. A total of 301 A. baumannii isolates, including 189 CRAB and 112 carbapenem-susceptible A. baumannii (CSAB), generated 602 high-quality spectra. Train–test splitting was performed at the isolate level to prevent information leakage. Fixed binning and K-means clustering–based dynamic binning strategies were compared, with K-means-derived bin boundaries learned from the training set and fixed for independent test-set transformation. Logistic regression, support vector machine, random forest, Light Gradient Boosting Machine, and Extreme Gradient Boosting models were trained and optimized using feature selection uniformly performed on the entire training dataset prior to nested cross-validation for hyperparameter optimization, and their performance was evaluated on an independent test set. The optimal model was interpreted using SHapley Additive exPlanations (SHAP).

Results

No significant differences were observed between CRAB and CSAB in clinical sources or baseline characteristics. Dynamic binning improved predictive performance across models. Optimal performance was achieved with 650 bins and 50 retained features. Under this setting, Light Gradient Boosting Machine (LightGBM) achieved an area under the receiver operating characteristic curve (ROC-AUC) of 0.985 and an accuracy of 0.942 on the test set. SHAP analysis identified multiple m/z intervals and statistical features that contributed to CRAB discrimination.

Conclusions

This study presents a rapid CRAB detection strategy based on MALDI-TOF MS and machine learning. Using dynamic binning and optimized feature selection, the LightGBM model achieved excellent predictive performance, with a ROC-AUC of 0.985 and an accuracy of 0.942 on the independent test set. The framework requires no additional experimental procedures, provides rapid predictions, and offers enhanced interpretability through SHAP analysis. The proposed framework warrants further validation in clinical settings for rapid antimicrobial resistance screening.