Prediction of methicillin-resistant Staphylococcus aureus and carbapenem-resistant Klebsiella pneumoniae from Raman spectra by Artificial Intelligent Raman Detection and Identification System (AIRDIS) with machine learning.

Lin, Hsiu-HsienHsiu-HsienLinLin, Yu-TzuYu-TzuLinChen, Chih-HaoChih-HaoChenTseng, Kun-HaoKun-HaoTsengHsu, Pang-ChienPang-ChienHsuWu, Ya-LunYa-LunWuChang, Wei-ChengWei-ChengChangLiao, Nai-ShunNai-ShunLiaoChou, Yi-FanYi-FanChouHsu, Chun-YiChun-YiHsuLiao, Yu-HuiYu-HuiLiaoHo, Mao-WangMao-WangHoChang, Shih-ShengShih-ShengChangPO-REN HSUEHCho, Der-YangDer-YangCho2025-08-272025-08-272025-07-3009248579https://scholars.lib.ntu.edu.tw/handle/123456789/731638Objectives: Methicillin-resistant Staphylococcus aureus (MRSA) and carbapenem-resistant Klebsiella pneumoniae (CRKP) are two of the most important antibiotic-resistant bacteria. Early use of the correct treatment strategy can not only reduce patient mortality but also prevent the development of resistance. Although some rapid but costly techniques are available, routine workflows in clinical microbiology laboratories can sometimes take several days to deliver bacterial identification and resistance profiles. In this study, we developed a bacterial identification and resistance prediction system that combines Raman spectroscopy and machine learning to predict the MRSA and CRKP. Methods: A total of 988 S. aureus isolates (including 513 MRSA) and 1053 K. pneumoniae isolates (including 517 CRKP) were collected. Of these, 266 S. aureus isolates and 285 K. pneumoniae isolates were used for training, while the remainder were used for validation. Results: The system demonstrated high predictive performance, with accuracy and area under receiver operating characteristic (AUROC) values of 88% and 0.92 for MRSA prediction and 87% and 0.96 for CRKP prediction, respectively. Conclusions: As a result, we confirmed the ability of machine learning to interpret Raman spectra for predicting resistant bacteria in clinical microbiology laboratories. This is the first and novel system validated with a large number of clinical isolates and may be incorporated into existing workflows.enfalseArtificial IntelligenceCarbapenem-resistant Klebsiella pneumoniaeMachine learningMethicillin-resistant Staphylococcus aureusPredictionRaman spectraPrediction of methicillin-resistant Staphylococcus aureus and carbapenem-resistant Klebsiella pneumoniae from Raman spectra by Artificial Intelligent Raman Detection and Identification System (AIRDIS) with machine learning.journal article10.1016/j.ijantimicag.2025.107587407500442-s2.0-105013189789