A probabilistic approach to the analysis of elastic light scatter profiles for identification of culturable bacteria

Abstract

Elastic Light Scatter (ELS) profiling is a unique, non-invasive technique for the rapid analysis and identification of bacterial colonies grown on agar. Colonies are inspected using a laser; the resulting light-scatter images have been shown to be species-specific. We examined a probabilistic approach to the analysis of ELS profiles for identification purposes. We analyzed 1701 colonies representing 52 strains across 19 food-related bacterial species representing four phylogenetically distinct families. Each of three ELS-derived feature sets (Patsekin elements, Zernike moments, pseudo-Zernike moments) were included, yielding a total of 2257 features. Each individual feature was binarized using an adaptive threshold set at 50% of its maximum value, a strategy that preserved meaningful differences across feature types of varied scale. A species-level identification matrix was constructed by summarizing positively expressed features across colonies, and representative species profiles were generated for comparison. Several taxa sharing ecological niches and ELS features were combined for improved performance. Use of Hypothetical Median Organism – and Leave-one-out analyses subsequently correctly identified each of 15 defined taxa with high Willcox probability scores. Importantly, test species not included in the database (Arcobacter bilvalviorum, A. faecis, and A. ellisii) were correctly considered as unknown based on a combined analysis of probabilistic scores and taxonomic distances, even when profiles shared partial similarity. A probabilistic approach to identifying bacteria from ELS profiles is an effective means to achieve this goal, and could potentially be used for other methods that generate taxon-specific spectra.