OBJECTIVE To screen characteristic peaks of hypervirulent Klebsiella pneumoniae(hvKP)using ma-trix-assisted laser desorption ionization time-of-flight mass spectrometry(MALDI-TOF MS)combined with EX-Smartspec software and establish a rapid detection model for hvKP.METHODS Based on identification criteria of any positive peg-344,iroB,iucA,rmpA,prmpA2 genes or siderophore production＞30 μg/ml,89 hvKP and 72 classical Klebsiella pneumoniae(cKP)strains were initially collected and validated for virulence via Galleria mellonella assays.A diagnostic model distinguishing hvKP from cKP was constructed using EX-Smartspec soft-ware and a convolutional neural network algorithm,integrating characteristic peaks and cluster analysis to provide a rapid and accurate clinical diagnostic tool.RESULTS MALDI-TOF MS analysis identified a characteristic hvKP peak at(3 835±100)ppm.Receiver operating characteristic(ROC)curve analysis revealed optimal performance in distinguishing hvKP with an area under the curve(AUC)=0.741.When AUC ≥0.089,the model demonstra-ted high sensitivity(86.41％),specificity(69.90％),accuracy(78.16％),positive predictive value(74.17％),and negative predictive value(83.72％)in differentiating hvKP from cKP.Cluster analysis further validated the model's classification accuracy.Additionally,the typing classification model exhibited high accuracy(approxi-mately 0.95 and 0.90 in training and validation phases,respectively)and low loss values(-0.18 and 0.30).Val-idation of 6 randomly selected hvKP and 5 cKP strains showed a 100.00％pass rate.CONCLUSION The estab-lished diagnostic model for hvKP and cKP provides a rapid and accurate clinical tool for timely treatment of hvKP-related infections.

OBJECTIVE To screen characteristic peaks of hypervirulent Klebsiella pneumoniae(hvKP)using ma-trix-assisted laser desorption ionization time-of-flight mass spectrometry(MALDI-TOF MS)combined with EX-Smartspec software and establish a rapid detection model for hvKP.METHODS Based on identification criteria of any positive peg-344,iroB,iucA,rmpA,prmpA2 genes or siderophore production＞30 μg/ml,89 hvKP and 72 classical Klebsiella pneumoniae(cKP)strains were initially collected and validated for virulence via Galleria mellonella assays.A diagnostic model distinguishing hvKP from cKP was constructed using EX-Smartspec soft-ware and a convolutional neural network algorithm,integrating characteristic peaks and cluster analysis to provide a rapid and accurate clinical diagnostic tool.RESULTS MALDI-TOF MS analysis identified a characteristic hvKP peak at(3 835±100)ppm.Receiver operating characteristic(ROC)curve analysis revealed optimal performance in distinguishing hvKP with an area under the curve(AUC)=0.741.When AUC ≥0.089,the model demonstra-ted high sensitivity(86.41％),specificity(69.90％),accuracy(78.16％),positive predictive value(74.17％),and negative predictive value(83.72％)in differentiating hvKP from cKP.Cluster analysis further validated the model's classification accuracy.Additionally,the typing classification model exhibited high accuracy(approxi-mately 0.95 and 0.90 in training and validation phases,respectively)and low loss values(-0.18 and 0.30).Val-idation of 6 randomly selected hvKP and 5 cKP strains showed a 100.00％pass rate.CONCLUSION The estab-lished diagnostic model for hvKP and cKP provides a rapid and accurate clinical tool for timely treatment of hvKP-related infections.