L'IA peut-elle prédire avec une précision de 99 % la probabilité qu'un individu développe une maladie génétique en utilisant uniquement l'analyse par IA de son microbiote et des données d'exposition environnementale ?

La prédiction génomique a progressé, mais les interactions environnementales restent mal modélisées. Les lois sur la vie privée et les préoccupations éthiques retardent l'adoption généralisée des prévisions au niveau individuel sans validation clinique.

Background

Genomic prediction has advanced, but environmental interactions remain poorly modeled; privacy laws and ethical concerns delay widespread individual-level forecasting without clinical validation.

As of 2024, AI can predict polygenic risks for a handful of common conditions (e.g., type 2 diabetes, colorectal cancer) by combining microbiome profiles with lifestyle and environmental data, but the models currently reach at best modest-to-moderate discrimination (AUC ≈ 0.65–0.80) rather than the claimed 99 % accuracy. Large consortia such as the American Gut Project and the UK Biobank have demonstrated that microbiome and exposome features explain only a small fraction of heritable genetic disease variance, and these models remain far from clinical-grade single-patient risk stratification. Integrating polygenic scores with transcriptomic or proteomic readouts further improves area-under-the-curve, yet the highest reported performances still fall well below 99 %. Demonstrating 99 % predictive accuracy for individual genetic-disease onset using only microbiome and environmental data has not been achieved and is not consistent with current heritability estimates.

— Enriched May 10, 2026 · Source: NIH Human Microbiome Project

While AI has made significant progress in analyzing microbiome and environmental exposure data to predict disease risk, predicting an individual's likelihood of developing any genetic disease with 99% accuracy remains an elusive goal. Current AI models can identify associations between certain microbiome patterns and disease risk, but they are not yet capable of achieving such high accuracy due to the complex interplay between genetic, environmental, and lifestyle factors. The current state of the art involves using machine learning models to identify high-risk individuals, but these models are often limited by the quality and quantity of available data, as well as the lack of a comprehensive understanding of the underlying biological mechanisms. As a result, AI-based predictions are typically used in conjunction with other diagnostic tools and clinical expertise to provide more accurate assessments.

— Status checked on May 10, 2026.