Researchers at the University of Auckland propose training AI on stool residuals from New Zealand's bowel screening programme to predict individual bowel cancer risk with a target accuracy of up to 90 percent, according to RNZ. Dr. Theo Portlock of the Liggins Institute says the AI would analyse gut microbiome bacteria in discarded Fecal Immunochemical Test (FIT) samples, modelling complex multi-species relationships that simpler analysis cannot detect. New Zealand has the fastest rate of early-onset bowel cancer in the world, and the project is currently seeking funding.
The key architecture insight is an augmentation-over-replacement pattern: FIT test kits already process stool samples across New Zealand's national bowel screening programme, and the residual material is currently discarded. Researchers are proposing to bolt a secondary microbiome classifier onto the end of that existing pipeline – no new patient enrollment, no new sample collection – to convert a single binary blood-detection signal into a multi-dimensional cancer risk score. For data practitioners, this is a textbook example of extracting latent signal from infrastructure that is already running.
Research fellow Dr. Theo Portlock at the University of Auckland's Liggins Institute is proposing to train AI models on the bacteria found in discarded FIT test residuals from New Zealand's National Bowel Screening Programme, per RNZ. FIT tests detect traces of blood in stool as a risk marker for bowel cancer. After testing, residual material is currently discarded. The proposal would sequence the gut microbiome from those samples and train an AI classifier to distinguish between individuals with bowel cancer and those whose blood traces are attributable to other causes, targeting up to 90 percent risk-prediction accuracy (RNZ).
The ML challenge is multi-feature, nonlinear pattern recognition across dozens of microbial taxa, where the signal may involve one species increasing, another decreasing, or more complex co-occurrence patterns. Portlock describes the approach: "Sometimes you might have an increase in one or a decrease in another set of protective species or even more complicated relationships. Now, AI is the only tool in our scientific toolbox that is able to model these without having anything predefined" (RNZ). This is characteristic of ensemble or attention-based classifiers rather than logistic regression. The specific model architecture, sequencing method (16S rRNA vs. shotgun metagenomics), and validation protocol are not detailed in the RNZ report, so the 90 percent accuracy headline cannot yet be assessed in terms of precision-recall balance or cohort size.
New Zealand has the fastest rate of early-onset bowel cancer in the world, and the reason is unknown – researchers are investigating microplastics, nitrates and their effect on the gut microbiome, and lifestyle factors such as BMI and smoking (RNZ). A symptomatic screening programme extension, expected to roll out next year, would expand FIT testing to individuals presenting with bowel symptoms, potentially growing the training corpus faster. Portlock states that benefits would include "reduced false positives, reduced false negatives, and hopefully reduced waiting times for colonoscopies" (RNZ).
This is a pre-funding proposal, not a deployed system. Watch for a funded study design with a peer-reviewed validation dataset, a specified sensitivity-specificity tradeoff replacing the headline 90 percent figure, and whether the symptomatic-screening cohort (a different risk distribution from general-population screening) is used for validation or as a separate evaluation set.
Interesting pre-clinical research proposal: AI classifying gut microbiome data from existing FIT test residuals to stratify bowel cancer risk. Practitioner-relevant approach (augmenting an existing national screening pipeline) and significant NZ context (world's fastest early-onset rate). However, this is a pre-funding proposal with no published validation data, architecture, or cohort details, placing it in the solid-but-early research tier.
A 5-minute Tuesday brief on AI & data science. Curated, no fluff.
No spam. Privacy.
Practice with real Health & Insurance data
90 SQL & Python problems · 15 industry datasets
250 free problems · No credit card
News on Let's Data Science is compiled from multiple public sources with editorial oversight. See our Editorial Standards and Corrections Policy.
Leave a Reply