# Near-Infrared Spectroscopy and Multivariate Analysis as an Effective Method to Discriminate Escherichia coli in Clinical Samples

## TL;DR

This study shows that near-infrared spectroscopy and multivariate analysis can accurately identify antibiotic-sensitive and -resistant Escherichia coli strains in clinical samples.

## Contribution

The novel use of SPA-LDA and GA-LDA models with NIR spectroscopy for rapid and accurate discrimination of E. coli strains.

## Key findings

- SPA-LDA and GA-LDA models achieved 100% sensitivity and specificity for E. coli strain discrimination.
- Spectral preprocessing techniques like Savitzky–Golay and EMSC improved data quality.
- NIR spectroscopy combined with variable selection is a promising tool for microbiological diagnostics.

## Abstract

Escherichia
coli is a bacterium
that inhabits the gastrointestinal system and is considered to be
an essential part of the intestinal microbiota. However, some strains
can be pathogenic, causing urinary tract infections. These bacteria
can develop antibiotic resistance during prolonged or inadequate treatments,
and sensitive and specific tests are necessary for diagnosis. In this
study, we aimed to determine, through NIR spectroscopy combined with
variable selection techniques such as the successive projections algorithm
(SPA) and genetic algorithm (GA) integrated with linear discriminant
analysis (LDA), the discrimination of E. coli strains (sensitive vs resistant). The two E. coli strains resulted in a total of 162 spectral data, classified into
81 sensitive and 81 resistant spectra. These data were later subdivided
into 114 for training, 24 for validation, and 24 for testing. Each
of these sets maintained a balanced proportion between the two strains,
containing half of the sensitive and half of the resistant strains.
The variables selected by these methods were used to differentiate
the species. Additionally, we evaluated the influence of spectral
preprocessing techniques such as Savitzky–Golay smoothing and
extended multiplicative scatter correction (EMSC) on the spectral
data. The results showed that both models (SPA-LDA and GA-LDA) presented
100% sensitivity and specificity for both sensitive and resistant
strains. This demonstrates that NIR spectroscopy combined with variable
selection techniques can be an effective method for rapid and accurate
identification of bacterial strains, offering a promising alternative
for microbiological diagnostics.