Signature Infrared Bacteria Spectra Analyzed by an Advanced Integrative Computational Approach Developed for Identifying Bacteria Similarity

Infectious diseases have become one of the leading causes of global morbidity and mortality in human life. Therefore, bacteria identification plays a critical role to protect public health from infectious diseases. Optical spectroscopy has been widely used to identify bacteria by yielding unique signature spectra representing the structure and composition of the bacteria. Since bacteria exhibit very close sensing and analytical patterns, it is challenging to determine their distinctions. In this study, we introduce a novel integrative computational approach for analyzing infrared (IR) spectroscopy spectra to identify bacteria by measuring their similarities employing logarithmic-based wavelet features. A visual analytics tool is also developed to support an interactive visual analysis for characterizing bacteria features. Thirty-two signature IR spectra of Escherichia coli (E. coli) and forty spectra of Pseudomonas aeruginosa (P. aeruginosai) collected by an advanced fiber-optic Fourier transform IR (FO-FTIR) spectroscopy sensor system are used for validating the proposed computational approach. Three machine learning techniques are employed to differentiate the bacteria by measuring their performances. Implementing the presented approach, we identified the overall accuracy of 92.5% in classifying the bacteria with logistic regression. We also found that both sensitivity and specificity are significantly high, exceeding 92%.