Back

Agentic AI for automated hypothesis testing in Alzheimer's disease and related dementias

Saichandran, K. S.; Elzokm, K.; Guney, O. B.; Kolachalama, V. B.

2025-12-04 neurology
10.64898/2025.12.02.25341517 medRxiv
Show abstract

BackgroundAlzheimers disease and related dementias (ADRD) are multifactorial disorders driven by genetic, vascular, inflammatory, lifestyle, and biological factors interacting over decades, requiring population-level analyses. However, utilizing multimodal datasets is labor-intensive, impeding discovery and equity. Multi-agentic AI systems promise workflow automation but lack ADRD-specific adaptations. MethodsWe created a modular AI system with agents for data preparation, planning, analysis, critique, and summarization. Using data from the National Alzheimers Coordinating Center (NACC) (n=48,876), we tested 100 literature-derived ADRD hypotheses, subsampled cohorts (100-10,000 participants), and 100 non-ADRD controls via iterative loops. ResultsThe system verified 42.0{+/-}2.8 hypotheses by rejecting the null, found no evidence for 16.6{+/-}1.1, and deemed 41.4{+/-}2.3 not testable, with consistency in negated versions. Verification rates rose from 10.4 (n=100) to 35.4 (n=10,000); controls showed 85.2{+/-}0.6 non-testable. ConclusionsOur system illustrates a powerful paradigm in which AI can facilitate automated data analysis, expediting advancements in population-level dementia investigations.

Matching journals

The top 9 journals account for 50% of the predicted probability mass.

50% of probability mass above

"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.