Advanced Statistical Analysis Enables Food Bacteria Detection

Salmonella typhimurium, in red, invading cultured human cells. (Rocky Mountain Laboratories, NIAID, NIH)

Finding bacterial contamination in the food supply when the target contaminants are known is difficult enough, but finding food-borne bacteria that are not identified targets can seem like an overwhelming task. To attack this problem, researchers from Indiana University-Purdue University Indianapolis (IUPUI) and Purdue University in West Lafayette, Indiana have developed a new approach to automated detection and classification of harmful bacteria in food.

The investigators designed and implemented a sophisticated statistical approach that allows computers to learn about new pathogens while the testing is underway, rather than dealing with a single set of known contaminants. This more adaptive approach can help address issues like the high mutation rate of bacteria, which multiplies the already high number of known pathogens.

Detecting pathogens such as listeria, staphylococcus, salmonella, vibrio, and E. coli was based on the optical properties of their colonies. For this job, the researchers used a prototype laser scanner, developed at Purdue University. The light-scattering patterns from the laser helped locate the bacterial colonies, with their identification and classification generated by the machine-learning capabilities of the computer system processing the data. For those capabilities, the team developed a Bayesian statistical approach to learning, starting with a non-exhaustive training dataset, which enabled the automated detection of unknown bacterial sub-species.

Murat Dundar, assistant professor of computer science at IUPUI and the university’s principal investigator called the findings a “step towards a fully automated identification of known and emerging pathogens in real time, hopefully circumventing full-blown, food-borne illness outbreaks in the near future.” The study appears in the October issue of the journal Statistical Analysis and Data Mining (paid subscription required).