基於質譜儀資料利用人工智慧方法辨識革蘭氏陰性菌對環丙沙星抗藥性之特徵峰值;Identification of Informative Peaks for Ciprofloxacin-resistant Gram-negative Bacteria Based on MALDI-TOF MS Using Artificial Intelligence Approaches

NCUIR > College of Electrical Engineering & Computer Science > Graduate Institute of Computer Science and Information Engineering > Electronic Thesis & Dissertation > Item 987654321/86525

Please use this identifier to cite or link to this item: http://ir.lib.ncu.edu.tw/handle/987654321/86525

Title:	基於質譜儀資料利用人工智慧方法辨識革蘭氏陰性菌對環丙沙星抗藥性之特徵峰值;Identification of Informative Peaks for Ciprofloxacin-resistant Gram-negative Bacteria Based on MALDI-TOF MS Using Artificial Intelligence Approaches
Authors:	張文睿;Zhang, Wen-Rui
Contributors:	資訊工程學系
Keywords:	基質輔助雷射脫附電離飛行時間質譜法;人工智慧;抗生素抗藥性;MALDI-TOF MS;artificial intelligence approach;antibiotic resistance
Date:	2021-07-23
Issue Date:	2021-12-07 12:56:19 (UTC+8)
Publisher:	國立中央大學
Abstract:	抗生素耐藥性已成為全球重要且日益嚴重的威脅，耐藥病原體感染會造成大量發病率和死亡率對患者構成威脅。基質輔助雷射脫附電離飛行時間質譜法（Matrix-assisted Laser Desorption Ionization-Time of Flight Mass Spectrometry MALDI-TOF MS）為一種新興的微生物分析技術，被廣泛應用於微生物菌種之鑑定，近年來亦有許多研究用以辨識細菌之抗藥性。目前文獻上的研究侷限於使用質譜資料來辨識某一革蘭氏陰性菌或革蘭氏陽性菌對環丙沙星(Ciprofloxacin)之抗藥性，仍少有使用質譜來辨識多種革蘭氏陰性菌對環丙沙星抗藥性特徵峰值之研究。本研究藉由長庚醫院蒐集的大量質譜資料，並針對四種革蘭氏陰性菌: Escherichia coli、 Klebsiella pneumoniae、 Acinetobacter baumannii 、與Acinetobacter nosocomialis，採用兩種特徵擷取方式並搭配隨機森林，卷積神經網路建構抗藥性之模型，接著再進一步分析不同特徵選取方式對於四種革蘭氏陰性菌對環丙沙星之抗藥性特徵峰值與其分布差異。在各個細菌中，皆以卷積神經網路建構之模型有較高之準確率，其在獨立測試的準確率分別為76.58%，75.49%，85.78%，88.52%。此外，發現峰值2918、4244、7928、8553 m/z重複出現在A. baumannii 和A. nosocomialis中，峰值3580、3926、6327 m/z 重複出現在A. baumannii 和E.coli中，峰值8351 m/z 重複出現在A. nosocomialis 和E.coli中，以及在K. pneumoniae中的峰值4570、5410、6153、7705 m/z出現在其他細菌上。對於A. baumannii有最多8個峰值在其他革蘭氏陰性菌上出現;但沒有存在三種或所有革蘭氏陰性菌皆出現之特徵峰值，這表示了在環丙沙星作用機制下，四種革蘭氏陰性菌仍保留其獨特峰值。本研究四種革蘭氏陰性菌特徵峰值分析可找尋對應蛋白質片段，以探究抗藥之原因。 ;MALDI-TOF MS is an emerging microbial analysis technology, which is widely used in the identification of bacterial species. However, there is lacking research to identify and analyze the informative peak of ciprofloxacin resistance Gram-negative bacteria. In this study, we col-lected a lot of mass spectrometry data and considered four Gram-negative bacteria (Esche-richia coli, Klebsiella pneumoniae, Acinetobacter baumannii, Acinetobacter nosocomialis), using two feature extraction methods, binning method, and kernel density estimation to process the MS data. After feature extraction, we constructed random forest and convolutional neural network model to predict antibiotic resistance, and then further analyze the informative peaks and distribution. In independent test, the binning method and convolutional neural network model has the highest accuracy 76.58%, 75.49%, 85.78%, and 88.52% for E. coli, K. pneu-moniae, A. baumannii and A. nosocomialis respectively. After feature selection, it was found that the peaks 2918, 4244, 7928, and 8553 m/z repeatedly appeared in A. baumannii and A. nosocomialis, the peaks 3580, 3926, and 6327 m/z repeatedly appeared in A. baumannii and E. coli, the peak 8351 m/z repeatedly appeared in A. nosocomialis and E. coli. And the peaks of 4570, 5410, 6153, and 7705 m/z of K. pneumoniae appeared on other bacteria. For A. baumannii, there are most 8 informative peaks appearing on the other three Gram-negative bacteria. However, there are no exist informative peaks that appear in three or all Gram-negative bacteria, which means that under the mechanism of action on ciprofloxacin, the four Gram-negative bacteria retain their unique peaks. In short, the analysis of informative peaks provide a worthy field to research further.
Appears in Collections:	[Graduate Institute of Computer Science and Information Engineering] Electronic Thesis & Dissertation

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	54	View/Open

社群 sharing

Loading...