Scientists at a university in Scotland have developed a technique which could help to identify the source of food poisoning in a better way than current methods.

The technique relies on a new machine learning method – known as the Minimal Multilocus Distance (MMD) method – which can be used to train a computer to identify likely sources with high accuracy.

It has been demonstrated at a theoretical level to attribute human cases to source reservoirs such as chicken, cattle and sheep but further research is required to build it into technology that could be used by health protection professionals.

Researchers at the University of Aberdeen showed the technique could match Campylobacter and potentially other common foodborne pathogens more accurately to their source of origin within a reduced timeframe. 

Identifying likely sources of Campylobacter
Advances in Whole Genome Sequencing (WGS) mean that the complete DNA sequence of an organism’s genome can be obtained at a single time. However, methods that efficiently mine these data are yet to be developed.

The study was led by Francisco Perez Reche and professor Norval Strachan, from the University of Aberdeen’s departments of Physics and Biological Sciences.

Perez Reche said when dealing with an outbreak, speed and accuracy of identifying the likely source is key.

“There are a number of existing methods to calculate the likely source of an infection but in order to work effectively, they either use only part of the genome sequencing, meaning results are less targeted, or if they use the whole genome, the calculations can take up to two days to perform. Our MMD method trains the computer to identify likely sources of origin of a Campylobacter infection within seconds,” he said.

Researchers proposed a fast method for source attribution which can deal with genotypes comprising thousands of loci with minimal computational effort.

Method accuracy
Whole genome sequenced Campylobacter isolates including 500 clinical isolates from human patients and 673 from five food and animal sources were obtained: 150 from cattle, sheep and chicken, 130 from pig and 93 from wild birds.

The total self-attribution accuracy when combining results across all the source populations was 73 percent for MMD and 57 percent for STRUCTURE, a Bayesian clustering model, in the Campylobacter example.

The MMD method correctly assigned most isolates (more than 70 percent) from pig, chicken and wild bird samples based on 25,937 core genome SNP (cgSNP) genotypes. Self-attribution of Campylobacter isolates from cattle and sheep is less precise at 58 percent and 45 percent. Wrongly self-attributed cattle isolates are mostly assigned to sheep and chicken sources, whilst sheep isolates tend to be erroneously attributed to cattle and chicken sources.

Source attribution was carried out to predict the origin of Campylobacter that resulted in human infection. MMD estimated that most cases (61 percent) were associated with chicken whilst wild birds and pigs were relatively unimportant (both less than 8 percent).

Researchers said it would be interesting to test accuracy of these methods for source attribution of human Campylobacter isolates from outbreaks with a known source.

Strachan added: “This has the potential to rapidly provide information on the potential source of infection and could be used to inform strategies to reduce food poisoning.”