At the inaugural Exposome Moonshot Forum, NEXUS had the opportunity to connect with diverse stakeholders from industry, academia, and government to discuss the future of the Human Exposome along with key The National Academies of Sciences, Engineering, and Medicine hosted a public webinar “Exploring the Types of Evidence Behind Diet and Chronic Disease” on July 10th 2025. During the webinar speakers explored the types of evidence used to look at relationships between diet and chronic disease.
NEXUS MPI Dr. Chirag Patel, is a Standing Member of the Committee on Evidence Synthesis and Communication in Diet and Chronic Disease Relationships. Dr. Patel spoke during this conference about “Using AI for Data Mining, Evidence Synthesis, and Data Collection” where he discussed the role of AI in sifting through evidence, particularly for modifiable factors of the exposome, such as nutrients from food.
Currently, before the advent of systematic measurement tools, such as those that are being brought to practice by the NEXUS team, sifting through data and studies can be daunting. Traditional evidence synthesis, like systematic reviews, is time-consuming and prone to biases. AI, especially large language models (LLMs) like ChatGPT and biomedical systems, could change that. Drawing from Dr. Patel’s presentation slides, emphasizing why a domain-wide or exposomics-wide approach is vital for reproducible exposomics research.
AI in Medicine: From Diagnosis to Discovery
In his presentation, Dr. Patel outlined AI’s dual roles: supervised AI for accurate diagnoses (e.g., a 2016 JAMA study on detecting diabetic retinopathy) and generative AI for innovations like drug proposals or virtual consultations. In precision nutrition, an example of supervised AI, a 2015 Cell paper from Dr. Eran Segal and colleagues showed how personalized diets predict glycemic responses via microbiota, diet, and activity—highlighting interpersonal variability key to the foodome.
The “ChatGPT revolution” brings summarization, content creation, programming, and problem-solving, but risks hallucinations and false positives. The FDA uses AI like “Elsa” for faster reviews. Dr. Patel questioned LLMs’ causal mechanisms, training data, and abilities in retrieval, bias detection, quality assessment (e.g., randomized vs. observational), and reproducibility—areas improving but needing evaluation.
Showcases include:
- “Data-to-paper” (NEJM AI): Autonomous LLM generating traceable research papers from data.
- Med-PaLM M: Generalist AI excelling on biomedical tasks like question answering and genomics.
- “Otto-SR” (using GPT-4): Superhuman screening for systematic reviews, cutting the usual 62-week timeline.
- GPT-4 feedback on papers rivaling human reviewers.
He also nods to mapping “exposure-phenome associations,” directly relevant to exposomics.
Domain-Wide Approaches for Exposomics and Reproducibility in the AI era
Exposomics demands holistic views of interconnected exposures, not silos. AI thrives on benchmarks, but evidence synthesis faces bottlenecks like resource demands and biases. Can AI systems be taught about the exposome?
Other key members of the webinar panel included Jessica M. Franklin, Optum and Marta Guasch-Ferré, University of Copenhagen who discussed “Population Level: Big data, Real-World Evidence”; Robert M. Califf, Duke University School of Medicine, Daniel Raiten, National Institutes of Health, Office of Nutrition Research, John Ioannidis, Stanford University who discussed “Individual Level: Clinical Trials, Alternatives” ; Gary A. Churchill, The Jackson Laboratory, James Wells, Cincinnati Children’s Hospital Medical Center who discussed “ Animal Studies: Model Species, Human Lab Construct”
To view the full webinar please visit: https://www.nationalacademies.org/event/45192_07-2025_exploring-the-types-of-evidence-behind-diet-and-chronic-disease-a-webinar