Please use this identifier to cite or link to this item:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/84540
Title: | 微流道與微流井陣列之表面增強拉曼散射平台結合機器學習以進行抗生素輔助之快速細菌鑑別 The Microfluidic Microwell-array SERS Platform combined with Machine Learning for Antibiotic-assisted Rapid Bacteria Discrimination |
Authors: | Po-Hsuan Chao 趙伯宣 |
Advisor: | 黃念祖(Nien-Tsu Huang) |
Keyword: | 微流道,表面增強拉曼光譜,細菌鑑別,機器學習, Microfluidic,SERS,Bacteria discrimination,Machine learning, |
Publication Year : | 2022 |
Degree: | 碩士 |
Abstract: | 近年來,使用微流道技術之機器學習相關生醫研究與應用正迅速崛起。微流道能夠以高通量處理微量體積之生物樣本,並生成大量數據,有助於各種機器學習方法的穩固性與預測表現。以收集樣本化學資訊數據進行微流道結合機器學習之應用而言,表面增強拉曼散射(Surface-enhanced Raman scattering, SERS)能夠做為非常有用的檢測方法,且具備快速、靈敏與免標記之優勢。透過SERS光譜進行機器學習之細菌鑑別應用對於決定細菌感染使用之抗生素藥物有其益處,因同菌種的細菌間仍可能對特定抗生素的敏感性有所差異。在本篇論文中,我們展示一種使用微流道微流井陣列結合表面增強拉曼散射(Surface-enhanced Raman scattering, SERS)之檢測平台,其中此微流井陣列可在2平方公分面積內生成800個獨立樣本。我們採用此平台來研究不同機器學習方法對於細菌菌種鑑別的成效,包含隨機森林(RF)、支援向量機(SVM)、K近鄰演算法(KNN)、以及卷積神經網路(CNN)。我們觀察到不同細菌菌株的SERS光譜可以輕易鑑別,但是細菌光譜在取得抗藥性基因後仍可能保留相似光譜型態,因此對多數機器學習方法來說需要更多訓練資料才能鑑別。然而,我們也收集了細菌受到30分鐘抗生素刺激後的SERS光譜,這些抗生素刺激誘導的代謝擾動會改變細菌SERS光譜成更容易鑑別的型態,所有機器學習方法只需400筆抗生素誘導光譜的訓練資料就足以達成98% 準確度的預測。這些結果意味未來基於SERS之微流道技術之機器學習相關細菌應用中30分鐘抗生素刺激可以作為一種輔助方法。 The utilization of microfluidic techniques for machine learning (ML)-based biomedical research and applications has emerged rapidly in recent years. Microfluidics has the inherent features of handling tiny volumes of biological samples with high-throughput functionality, which can help generate massive biological sample data that benefit the robustness and prediction performance of various machine learning methods. To collect the data of sample chemical information for microfluidic-ML application, Surface-enhanced Raman scattering (SERS) can be a powerful detection method, which has the advantage of rapid, sensitive, and label-free detection. The application of ML bacteria discrimination by SERS spectra would benefit the determination of antibiotic treatments for bacterial infections, as the bacteria of the same species could still have different susceptibilities to the certain antibiotic. Here, we demonstrate a microfluidic-microwell-array SERS platform that could generate 800 independent bacteria sample points in less than a 2 cm2 area. The platform is adopted to study the performances of various ML methods for bacteria strain discrimination, including random forest (RF), support vector machine (SVM), k-nearest neighbors (KNN), and convolutional neural network (CNN). We observe that the SERS spectra of several E. coli strains could be easily discriminated, but the spectra of the bacteria could remain similar patterns after gaining the antibiotic-resistant gene, thus requiring more training data for most ML methods to discriminate. However, we also collected the SERS spectra of these bacteria with 30-minute antibiotic incubation. The antibiotic-induced metabolic perturbations altered the SERS spectra to more distinguishable patterns, and the predictions by all ML methods of these antibiotic-induced spectra were able to reach 98% accuracy by only 400 training data. These results suggest that 30-minute antibiotic treatment can be an assisted approach for future SERS-based microfluidic-ML bacteria applications. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/84540 |
DOI: | 10.6342/NTU202203737 |
Fulltext Rights: | 同意授權(限校園內公開) |
metadata.dc.date.embargo-lift: | 2022-09-26 |
Appears in Collections: | 生醫電子與資訊學研究所 |
Files in This Item:
File | Size | Format | |
---|---|---|---|
U0001-2109202215505400.pdf Access limited in NTU ip range | 3.7 MB | Adobe PDF | View/Open |
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.