Standardization and evaluation of artificial intelligence-driven Fourier-transform infrared bio-typing for the discrimination of Escherichia coli and Shigella spp. on genus, species and serogroup level.

PURPOSE: Despite being genetically very closely related and sharing many phenotypical characteristics, Escherichia coli and Shigella spp. are regarded as distinct entities in epidemiology and clinical practice. Rapid and accurate discrimination is crucial for infection prevention and control and for surveillance. Presently they are distinguishable by laborious and time-consuming laboratory methods only. This study addressed the challenge of differentiating Escherichia coli and Shigella species by combining Fourier-Transform Infrared spectroscopy (FTIR) with machine learning as an automated typing method.

METHODS

Multicentric international collections comprising N = 2594 isolates, characterized at different levels (from full genome sequencing to mere species-level identification), were included in this study. Infrared spectra from water-ethanol bacterial suspensions of overnight cultures were acquired, processed and analysed using the FTIR-based IR Biotyper® system (IRBT; Bruker Daltonics GmbH & Co. KG, Germany). Three classifiers were created to distinguish E. coli from Shigella spp. (binary-classifier A, and at species level - classifier B) and to distinguish the most common STEC serogroups (classifier C). The classifiers were evaluated in terms of accuracy, error rate and failed classification rate.

RESULTS

The collection included n = 2334 E. coli and n = 260 Shigella spp. from human, animal, and food sources, resulting in a total of N = 19,824 spectra. Accuracy for classifier A was 2186/2189 (99.86%) and for classifier B 2185/2189 (99.82%). Errors were restricted to two rare Shigella spp. serotypes (for both classifiers) and one or two E. coli isolates for classifier A and B, respectively, one of which exhibiting a rough phenotype. Accuracy for classifier C was 1554/1554 (100%).

CONCLUSIONS

FTIR proved to be an accurate and robust technique for the discrimination of E. coli and Shigella species. The application of machine learning allows for automated data analysis and interpretation, reducing the likelihood of human bias. The IRBT system shows the potential of being applied as user-friendly and cost-effective method for routine workflows, after proper extensive validation.