Researchers at the University of Auckland are building an AI model that would analyze gut bacteria from discarded stool samples to predict bowel cancer risk, targeting an accuracy of up to 90 percent. The project, still seeking funding, could help New Zealand - which has the fastest rising rate of early-onset bowel cancer globally - triage patients more accurately and reduce colonoscopy wait times.
How the AI would bolt onto existing screening
The proposal is to sequence the gut microbiome from the leftover material of Fecal Immunochemical Test (FIT) kits, which currently test for blood in stool and then discard the remainder. Dr. Theo Portlock of the Liggins Institute said the AI would model complex microbial relationships that simpler analysis misses. "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," he told RNZ.
For data practitioners, this represents a textbook augmentation-over-replacement pattern: a secondary classifier is bolted onto an infrastructure that is already running nationwide, converting a binary blood-detection signal into a multi-dimensional risk score without new patient enrollment or sample collection. It's a clear example of how AI for Healthcare can be integrated into existing clinical pipelines.
The machine learning problem
The core challenge is nonlinear pattern recognition across dozens of microbial taxa, where signals can involve co-occurrence shifts rather than simple abundance changes. Portlock's description - increases in some species, decreases in others, or more intricate interactions - points toward ensemble or attention-based classifiers rather than linear models. The RNZ report does not specify the sequencing method, model architecture, or validation protocol, so the 90 percent accuracy target cannot yet be evaluated for precision-recall balance or cohort size.
New Zealand's urgent need
Bowel cancer incidence among younger populations is rising faster in New Zealand than anywhere else in the world, and the cause remains unclear. Researchers are investigating microplastics, nitrates, and lifestyle factors. A planned extension of the national screening programme to symptomatic patients next year could provide a larger training corpus. Portlock said that improved risk prediction would mean "reduced false positives, reduced false negatives, and hopefully reduced waiting times for colonoscopies."
What to watch
This is a pre-funding proposal, not a deployed system. Observers should watch for a funded study design with a peer-reviewed validation dataset, a specified sensitivity-specificity tradeoff to replace the headline 90 percent figure, and whether the symptomatic-screening cohort is used as a separate evaluation set or folded into training. The project's success will hinge on rigorous validation and transparent reporting of model performance.
Why this matters for Science & Research professionals
The project exemplifies a growing trend in AI for Science & Research: extracting latent signal from routine data collection that is already running, sidestepping the cost and time of recruiting new cohorts. The approach turns a single-threshold biomarker into a multi-dimensional risk score, and the technical challenge - modeling complex, non-linear interspecies relationships - mirrors many other problems in environmental and biological monitoring. The key takeaway is that infrastructure-first design can create high-value training corpora, but the claimed accuracy figure should remain bracketed until validation details are public.
Your membership also unlocks:
