dc.description.abstract | Matrix-assisted laser desorption ionization-time of flight mass spectrometry (MALDI-TOF MS) is a widely adopted method for bacteria species identification and also be reported can determinate the antibiotic resistance and reducing the time antibiotics susceptibility test (AST) cost. However, there has a lack of research for identifying and analyzing antibiotic-resistant Escherichia coli (E. coli), a common and well-known bacterium, by MALDI-TOF MS. Therefore, we construct machine learning models for E. coli antibiotic resistance prediction based on the MS data provided by Chang Gung Memorial Hospital. In our study, we build an alignment algorithm to handle the MS shifting problem and peak detection. We performed the two stage model training: split the dataset and build models for predicting resistance of five antibiotics, amoxicillin, ceftazidime, ciprofloxacin, ceftriaxone and cefuroxime by using logistic regression, support vector machine, random forest and eXtreme Gradient Boosting (XGBoost). XGBoost has the best performance in four machine learning models and achieve 0.62, 0.72, 0.87, 0.72, 0.72 AUC respectively in the subset which presents informative peak m/z 4533 for AMC and m/z 9714 for the rest four antibiotics. Furthermore, we identified the informative peaks for resistance determination of five antibiotics. In short, we build the antibiotic resistance prediction models of E. coli which can support doctor to make medication decision, also the analysis of informative peaks provide a worthy field to research further. | en_US |