Researchers at the University of Chicago have developed an AI-driven “digital twin” microbiome model named Q-net, which uses data from fecal samples of preterm infants in neonatal intensive care units (NICUs) to predict cognitive deficits with 76% accuracy.
While the gut microbiome significantly influences infants’ health and development, traditional lab experiments are insufficient to fully understand bacterial interactions and their potential to cause gastrointestinal diseases and neurodevelopmental issues.
Ishanu Chattopadhyay, PhD, assistant professor of medicine, noted that studying microbiome snapshots has limitations due to the constantly changing nature of preterm infants’ microbiomes. Consequently, the team adopted a generative AI approach to create a digital twin, simulating bacterial interactions as they evolve.
The digital twin technology represents complex systems digitally, enabling high-fidelity simulations of perturbations and failures that are impractical to replicate in reality, as explained in Science Advances. This comprehensive simulation contrasts with the typically narrow focus of standard machine learning models.
In contrast to wet lab experiments, which could take millennia to explore bacterial interactions in detail, the Q-net model accomplishes this much faster. Data from 58 infant fecal samples at UChicago’s Comer Children’s Hospital trained the model, while data from 30 preterm infants at Beth Israel Deaconess Medical Center in Boston validated its predictions.
The model demonstrated 76% accuracy in forecasting cognitive deficits, gauged by head circumference growth. Additionally, it suggested that restoring certain bacterial species could lower developmental risks for 45% of the infants, although it warned that incorrect interventions could exacerbate risks.
As reported by NutraIngredients, Chattopadhyay emphasized that simply administering probiotics is insufficient; the specific bacteria and timing are crucial. The Q-net model’s ability to pinpoint promising bacterial combinations can significantly streamline the search for effective interventions.