MIT Research Group Debuts FINGERS-7B as First Open-Source AI Foundation Model for Alzheimer’s Prevention

FINGERS-7B is the first AI foundation model for Alzheimer’s prevention, using multi-omic data to increase diagnostic accuracy by 4x. Open source at MIT.

By: AXL Media

Published: Apr 28, 2026, 6:14 AM EDT

Source: Information for this report was sourced from EurekAlert!

Engineering a Foundation for Early Neurodegenerative Detection

An international research consortium centered at MIT has released FINGERS-7B, marking the first AI foundation model specifically engineered for the prevention of Alzheimer’s disease. Presented at the ICLR conference in Rio de Janeiro on April 27, 2026, the model is designed to identify individuals at risk during the preclinical stage, which can precede cognitive decline by a decade. FINGERS-7B utilizes a massive dataset from tens of thousands of at-risk participants to refine its predictive capabilities. According to project co-lead Adrian Noriega, an MIT-Novo Nordisk AI Fellow, the system acts as a discovery engine that interprets complex biological signals to find novel interventions and therapeutics.

The Integration of Multi-Omic Biomarkers and Lifestyle Data

The technical innovation of FINGERS-7B lies in its ability to process diverse data streams simultaneously rather than analyzing single domains in isolation. The model learns from a combination of genomic, proteomic, and clinical signals alongside lifestyle factors such as diet and physical activity. This "multi-omic" approach allows the AI to detect subtle patterns that individual data sources might miss. By reading these biological domains together, the model provides a comprehensive "biological fingerprint" for each user, allowing for a more nuanced understanding of how disparate risk factors converge to trigger the onset of neurodegeneration.

Quantifying the Leap in Diagnostic Precision and Stratification

On the World-Wide FINGERS network datasets, the FINGERS-7B model has demonstrated significant performance improvements over prior technological standards. Research data indicates a 400% increase in the accuracy of preclinical diagnoses and a 130% improvement in responder stratification, which helps scientists identify which individuals are most likely to benefit from specific treatments. Li-Huei Tsai, director of the MIT Aging Brain Initiative, noted that the project overcomes the historical challenge of integrating vast volumes of genetic and epigenetic profile data into a single, actionable view of a patient’s prognosis.