What is it about?
This article reviews two papers that have come out recently that use machine learning to identify which food source someone with food poisoning may have got their infection from. They both focus on a common cause of food poisoning, salmonella, but use different approaches to building their models. While both models are quite accurate, one links many samples taken from humans back to a human source, suggesting no food poisoning occurred. A later algorithm doesn't do this, and suggests that the first algorithm overfit to its training data, giving us insight into some of the pitfalls of building machine learning algorithms for use in the real world.
Featured Image
Why is it important?
Machine learning is becoming more popular for predicting which bacteria will be a public health concern and which won't. We want these algorithms to be accurate and trustworthy, so understanding how they work and how they might make the wrong decisions is important.
Perspectives
This is the first example published where two groups have taken on the same problem in this field and the second group has done a comparison of the two models. This creates a huge learning opportunity for the community building these models to think critically about what their models are doing, and encourages healthy skepticism from people whose lives may be affected by the use of these models in public health in the future.
Dr Nicole E Wheeler
Wellcome Sanger Institute
Read the Original
This page is a summary of: Tracing outbreaks with machine learning, Nature Reviews Microbiology, February 2019, Springer Science + Business Media,
DOI: 10.1038/s41579-019-0153-1.
You can read the full text:
Contributors
The following have contributed to this page
