Accurate and effective discrimination of E. coli and Shigella is an important clinical issue, and there are many limitations in traditional methods of analysis. FT-IR shows great potential in the classification of bacteria with high specificity and low cost. In this study, we evaluated the efficiency of this technique when combined with multivariate analysis for rapid classification of E. coli and Shigella, which is difficult using traditional analytical methods. Machine learning and statistical tools were employed in combination with FT-IR to classify 14 E. coli and 9 Shigella strains. The classification accuracies for select E. coli and Shigella strains from blood agar were 0.7826, 0.8696, and 0.9565 at the genus, species, and strain levels, respectively. In addition, we used the FT-IR data of select strains from three different culture media for cross-validation, yielding an accuracy of 0.3681 at the strain level. These results indicate that the bacterial culture conditions have a significant impact on the FT-IR patterns. Based on this, an improved strategy for training an ensemble classifier model considering bacterial culture factors was constructed, resulting in almost perfect separation with an accuracy of 0.9394 for strain-level classification. These results show the potential of FT-IR combined with multivariate analysis for more reliable bacterial classification.
Keywords: Bacterial classification; FT-IR; Multivariate analysis; Support vector machine (SVM).
Copyright © 2022 Elsevier B.V. All rights reserved.