The scientists from Purdue University, Indianapolis, said the novel statistical method, which uses a laser scanner to analyse the optical characteristics of the bacteria, could be deployed in a network of labs and serve as a national bio-warning system to prevent foodborne illnesses. The method is outlined in the journal Statistical Analysis and Data Mining.
The team said the development of technologies for the rapid detection of bacterial pathogens was “crucial for securing the food supply
Statistical model
The investigators have designed an advance statistical model that boosts the ability of computers to detect the presence of bacterial contamination in tested samples. These formulas drive machine-learning, making possible the identification of known and unknown classes of food pathogens.
"The sheer number of existing bacterial pathogens and their high mutation rate makes it extremely difficult to automate their detection," said study leader M. Murat Dundar, Ph.D., assistant professor of computer science in the university’s School of Science. "There are thousands of different bacteria subtypes and you can't collect enough subsets to add to a computer's memory so it can identify them when it sees them in the future. Unless we enable our equipment to modify detection and identification based on what it has already seen, we may miss discovering isolated or even major outbreaks."
Combination
The team used a prototype laser scanner to spot and classify colonies of pathogens such as listeria, staphylococcus, salmonella, vibrio and E. coli based on the optical properties of their colonies. The success of the system rests on the combination of the new laser and the enhanced machine learning approach.
“Without the new enhanced machine-learning approach, the light-scattering sensor used for classification of bacteria is unable to detect classes of pathogens not explicitly programmed into the system's identification procedure”, said a statement from the Purdue scientists.
"We are very excited because this new machine-learning approach is a major 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,” said Dundar. “Ultimately we would like to see this deployed to tens of centres as part of a national bio-warning system.”
The team said its work was not based on any particular property of light scattering detection and therefore it could potentially be applied to other label-free techniques for classification of pathogenic bacteria, such as various forms of vibrational spectroscopy.
A Machine-Learning Approach to Detecting Unknown Bacterial Serovars by Ferit Akova1, Murat Dundar1, V. Jo Davisson, E. Daniel Hirleman, Arun K. Bhunia, J. Paul Robinson and Bartek Rajwa is published in October’s issue of Statistical Analysis and Data Mining; DOI:10.1002/sam.10085