Prism: ethically-guided multimodal AI models for predicting disease pathogenesis in individuals with pathogenic variants

Massachusetts General Hospital received a $100,000 Cooperative Agreement from the National Institutes of Health Office of the Director under the Trans-NIH Research Support program (CFDA 93.310), awarded September 2, 2025, with completion by August 31, 2027. The award supports development of ethically-guided machine learning and artificial intelligence (AI/ML) tools to predict disease manifestation and progression in individuals carrying pathogenic genetic variants, with initial focus on Alzheimer's disease and cardiomyopathy. The project integrates multimodal data from large biobanks and healthcare systems—including genomic, clinical, imaging, and other biomarkers—into unified computational models to improve risk stratification and identify disease biomarkers applicable to clinical practice. Complementing the technical development, the project incorporates research on the ethical, legal, and social implications (ELSI) of these AI tools in genomic medicine to ensure responsible implementation and mitigate potential biases across diverse populations.

The cooperative agreement engages two sub-awardees to deliver specialized components: the Broad Institute of MIT and Harvard contributes advanced AI/ML model development for cardiomyopathy prediction using multimodal datasets from FinnGen, UK Biobank, and Massachusetts General Hospital biobanks, with emphasis on contrastive learning frameworks and missing data imputation; and Cedars-Sinai Medical Center, led by Dr. Nicholas Tatonetti, develops phenotyping algorithms using electronic health record data and natural language processing, designs multimodal deep learning architectures integrating structured and unstructured clinical data, and ensures alignment with the project's ELSI framework through bias assessment and model interpretability techniques. Validated tools, best practices, and lessons learned will be disseminated to advance genomic risk assessment and management in clinical practice.

Abstract

This project aims to develop machine learning and artificial intelligence tools to enhance prediction of disease manifestation in individuals with pathogenic genetic variants. We will focus on alzheimer's disease and cardiomyopathy, leveraging diverse data from large biobanks and healthcare systems to create a more comprehensive and nuanced understanding of how genetic variants interact with other factors to influence disease onset and progression. Our interdisciplinary team will create an ethical framework to guide tool development and implementation, integrating genomic, clinical, imaging, and other data types into a common model. We will build and validate tools to predict disease onset, progression, and treatment response. We will then perform rigorous cross-validation to assess the generalizability of these tools across different datasets and clinical settings, ensuring their robustness and applicability in diverse healthcare contexts. Using a combination of advanced AI and interpretable machine learning methods, we aim to identify novel biomarkers for disease that could be more readily translated into clinical practice. Integrated with the technical development, we will conduct research on the ethical, legal, and social implications (elsi) of incorporating these tools into genomic medicine. The results of these elsi research projects will be further integrated into our ethical framework to inform the development of the optimal AI tools. This project will advance the field by improving risk stratification for variant carriers and identifying factors influencing disease beyond genetics alone. We aim to disseminate validated tools, best practices, and lessons learned to the broader research community, potentially transforming how genetic risk is assessed and managed in clinical practice. By combining cutting-edge computational methods with careful ethical consideration, we hope to create resources that can significantly enhance patient care and our understanding of complex genetic diseases. Ultimately, this work will lead to more targeted prevention strategies, personalized treatment plans, and improved outcomes for individuals at risk of alzheimer's disease, cardiomyopathy, and other genetically influenced conditions.