Detection of carbapenem-resistant Klebsiella pneumoniae on the basis of matrix-assisted laser desorption ionization time-of-flight mass spectrometry by using supervised machine learning approach.

BACKGROUND

Carbapenem-resistant Klebsiella pneumoniae (CRKP) is emerging as a significant pathogen causing healthcare-associated infections. Matrix-assisted laser desorption/ionisation mass spectrometry time-of-flight mass spectrometry (MALDI-TOF MS) is used by clinical microbiology laboratories to address the need for rapid, cost-effective and accurate identification of microorganisms. We evaluated application of machine learning methods for differentiation of drug resistant bacteria from susceptible ones directly using the profile spectra of whole cells MALDI-TOF MS in 46 CRKP and 49 CSKP isolates.

METHODS

We developed a two-step strategy for data preprocessing consisting of peak matching and a feature selection step before supervised machine learning analysis. Subsequently, five machine learning algorithms were used for classification.

RESULTS

Random forest (RF) outperformed other four algorithms. Using RF algorithm, we correctly identified 93% of the CRKP and 100% of the CSKP isolates with an overall classification accuracy rate of 97% when 80 peaks were selected as input features.

CONCLUSIONS

We conclude that CRKPs can be differentiated from CSKPs through RF analysis. We used direct colony method, and only one spectrum for an isolate for analysis, without modification of current protocol. This allows the technique to be easily incorporated into clinical practice in the future.