Pathogen identification with laser-induced breakdown spectroscopy: the effect of bacterial and biofluid specimen contamination

Abstract

In this paper, the potential use of laser-induced breakdown spectroscopy (LIBS) for the rapid discrimination and identification of bacterial pathogens in realistic clinical specimens is investigated. Specifically, the common problem of sample contamination was studied by creating mixed samples to investigate the effect that the presence of a second contaminant bacterium in the specimen had on the LIBS-based identification of the primary pathogen. Two closely related bacterial specimens, Escherichia coli strain ATCC 25922 and Enterobacter cloacae strain ATCC 13047, were mixed together in mixing fractions of 10:1, 100:1, and 1000:1. LIBS spectra from the three mixtures were reliably classified as the correct E. coli strain with 98.5% accuracy when all the mixtures were withheld from the training model and classified against spectra from pure specimens. To simulate a rapid test for the presence of urinary tract infection pathogens, LIBS spectra were obtained from specimens of Staphylococcus epidermidis obtained from distilled water and sterile urine. LIBS spectra from the urine-harvested bacteria were classified as S. epidermidis with 100% accuracy when classified using a model containing only spectra from other Staphylococci species and with 88.5% accuracy when a model containing five genera of bacteria was utilized. Bacterial specimens comprising five different genera and 13 classifiable taxonomic groups of species and strains were compiled in a library that was tested using external validation techniques. The importance of utilizing external validation techniques where the library is tested with data withheld from all previous testing and training of the model was revealed by comparing the results against "leave-one-out" cross-validation results. Last, the effect of using sequential models for the classification of a single unknown spectrum was investigated by comparing the misclassification of two closely related bacteria, E. coli and E. cloacae, when the classification was first performed using the five-genus bacterial library and then with a smaller model consisting only of E. coli and E. cloacae specimens. This result shows the utility of using successively more targeted analyses and models that use preliminary classifications from more general models as input.

A comparison of multivariate analysis techniques and variable selection strategies in a laser-induced breakdown spectroscopy bacterial classification

“…cultured in different media, grown on different days over the course of 18 months, exposed to different environmental stresses, etc.) [8] This is shown in Table 1.…”

“…[1][2][3][4] Over the last five years we and others have demonstrated a sensitive and specific identification of live bacterial biospecimens utilizing a discriminant function analysis (DFA) to classify LIBS spectra. [5][6][7][8] The intensities of strong specific elemental atomic emission lines normalized by the total observed spectral power have been utilized as independent variables in this multivariate analysis. [9] The selection of specific spectral lines to serve as independent variables in the multivariate analysis is known as variable down-selection.…”

“…Variables were down-selected from bacterial LIBS spectra obtained from five different genera and 13 distinct taxonomic classes of species and strains. [8] Model performance was quantified by calculating truth tables (and the resulting sensitivity and specificity) from the external validation tests.…”

Recent advances in the use of laser-induced breakdown spectroscopy (LIBS) as a rapid point-of-care pathogen diagnostic

“…The LIBS data are averaged relative emission intensities from 669 bacterial spectra acquired from bacteria with 13 different taxonomic classifications. 15 Approximately 32 different specimens (different cultures, tested over many months) are represented. 15 but element has been observed by others in bacterial LIBS spectra…”

Differentiation of fibrotic liver tissue using laser-induced breakdown spectroscopy

“…However, the use of internal standards could produce inconsistent results when water is present in the analyzed tissues [31]. Other works [22,32], implemented multivariate statistical techniques to distinguish between different samples. In particular, the application of Principal Component Analysis (PCA) has demonstrated its worth to find patterns in confusing data sets [33].…”