Skip to main content
NUTRITION 2026 Main banner

Artificial Intelligence and Machine Learning in Nutrition

Category: Artificial Intelligence and Machine Learning in Nutrition (Poster Session)

Artificial Intelligence and Machine Learning in Nutrition (Poster Session)

(P03-25-26) From Food Metabolome to Bioactivity: An Artificial Intelligence-Assisted Index Reveals Fruit-Specific Biological Activity Profiles

Saturday, July 25, 2026
11:30 AM - 12:30 PM ET
Location: Poster Board D-89

    Presenting Author(s)

  • TX

    Tong Xia, MD, PHD, MPH

    Postdoc research fellow
    Harvard Medical School and Brigham and Women's Hospital; Harvard School of Public Health
    Boston, Massachusetts, United States

    • Co-Author(s)

  • NN

    Nile Nair

    American Heart Association, United States

  • MS

    Michael L. Sebek

    Northeastern University, United States

  • SL

    Salvo D. Lombardo

    Harvard Medical School, United States

  • GI

    Gordana Ispirova

    Harvard Medical School, United States

  • SA

    Selena Ahmed

    American Heart Association, United States

  • ER

    Eric B. Rimm, ScD

    Professor
    Harvard T.H. Chan School of Public Health
    Boston, Massachusetts, United States

  • MS

    Meir J. Stampfer, MD, PhD

    Professor
    Channing Division of Network Medicine, Department of Medicine, Brigham and Women's Hospital and Harvard Medical School
    Boston, Massachusetts, United States

  • Frank B. Hu, MD, MPH, PhD (he/him/his)

    Chair, Department of Nutrition; Fredrick J. Stare Professor of Nutrition and Epidemiology
    Harvard T.H. Chan School of Public Health
    Boston, Massachusetts, United States

  • GM

    Giulia Menichetti

    Brigham and Women's Hospital, Massachusetts, United States

  • DH

    Danielle E. Haslam, PhD

    Instructor in Medicine
    Brigham and Women's Hospital and Harvard Medical School
    Canton, Massachusetts, United States

  • SB

    Shilpa N. Bhupathiraju, PhD (she/her/hers)

    Associate Professor in Medicine and Nutrition
    Harvard Medical School and Harvard T.H. Chan School of Public Health
    Boston, Massachusetts, United States

Disclosure(s):
Shilpa N. Bhupathiraju, PhD: Almond Board of California: Grant/Research Support (Ongoing)
California Walnut Commission: Grant/Research Support (Ongoing)
The Peanut Institute: Grant/Research Support (Ongoing)


Objectives: Fruits are among the most widely recommended foods for chronic disease prevention, yet no framework exists to quantify the biological activity of a whole fruit based on its full metabolite profile. We developed the Food Bioactivity Index (FBI), a novel framework integrating food metabolomics and artificial intelligence-assisted bioassay classification to map fruit biological activity across disease-relevant categories.

Methods: Untargeted metabolites from nine fruits (apple, orange, blueberry, strawberry, peach, grapes, cantaloupe, grapefruit, and apricot) were identified using the Periodic Table of Food Initiative and matched to experimental activity records in ChEMBL database. A large language model (OpenAI gpt-4o-mini) categorized 16,908 experiments into 13 disease-relevant activity domains (e.g., anticancer); a second model (Anthropic Claude Haiku) verified this independently (Cohen's κ = 0.64). For each fruit and activity domain, the FBI score is the abundance-weighted mean activity of its metabolites, weighting more abundant and more bioactive metabolites more heavily. A background test confirmed genuine enrichment against randomly sampled ChEMBL compounds.

Results: Identified metabolites ranged from 22 (blueberry) to 95 (grapes). After background correction, 11 of 13 activity domains were enriched. FBI scores were fruit-specific (range 4.09–7.89; higher = stronger activity), reflecting distinct experimental bioactivity domains of metabolites in each fruit: blueberry scored highest for antioxidant (7.89), anticancer (7.55), antiproliferative (6.72), antiviral (6.66), antidiabetic (6.44), and cardioprotective (5.51); apricot for antiaging (7.48) and antitumor (5.99); grapes for anti-inflammatory (6.43); and apple for antifungal (7.12) and antibacterial (6.32).

Conclusions: The FBI offers a whole-metabolome bioactivity map of commonly consumed fruits, revealing a distinct biological activity fingerprint for each fruit beyond known nutrients, with implications for understanding fruit-disease associations and guiding future human studies.

Funding Sources: AHA