Genomics-empowered AI for personalized cancer risk assessment, monitoring, and prevention

The National Human Genome Research Institute awarded Icahn School of Medicine at Mount Sinai a $4,643,754 cooperative agreement on April 1, 2026, under the Human Genome Research program (CFDA 93.172) to develop genomics-empowered artificial intelligence models for personalized cancer risk assessment, monitoring, and prevention.

The project, a MAGEN Development Site, aims to address the gap between theoretical potential and practical application of genomics-based AI in clinical and population healthcare by developing state-of-the-art AI models that accurately identify pathogenic variants affecting DNA repair genes and predict cancer risks over an individual's life course. The work will optimize screening and prevention strategies in an ethical, legal, and social implications (ELSI)-informed manner. Aim 1 focuses on developing robust genomic AI models for identifying protein-disrupting missense variants conferring high cancer risks. Aim 2 combines genetic factors, including common and rare variant polygenic risk scores, with non-genetic factors from electronic health records to predict disease risk across the life course.

The performance period runs April 1, 2026, through March 31, 2028, with work performed in New York. The multidisciplinary team comprises experts from Icahn School of Medicine at Mount Sinai, Boston Children's Hospital/Harvard, and Columbia University in computational genomics, AI/ML, health informatics, statistical genetics, medical genetics, population health, oncology, and ELSI research.

Boston Children's Hospital, awarded a sub-grant on June 30, 2026, will develop and validate risk models for breast, ovarian, and Lynch syndrome cancer genes, integrate additional risk variants and social determinants of health, collaborate on ELSI frameworks and focus groups, and develop educational strategies for communicating risk estimates across diverse health literacy levels. Columbia University, awarded a sub-grant on May 27, 2026, will formulate an ELSI framework for machine learning and AI in genomics and lead focus group planning, development, participant recruitment, moderation, and qualitative analysis under ELSI leadership.

Abstract

This proposed magen development site aims to develop genomics and multi-modal artificial intelligence (AI) models to transform personalized cancer risk assessment, monitoring, and prevention. A substantial GAP exists between the theoretical potential of genomics-based AI predictions and their practical application in clinical and population healthcare settings. The clinical classification of genetic variants is hindered by insufficient data to classify ultra-rare variants, particularly those found in non-european populations. Moreover, despite significant advances in AI across fields, we lack AI models that can combine diverse streams of health data to accurately predict disease risk across a person's life course. Finally, the real-world effectiveness of these AI models remains untested and their ethical, legal, and social implications (elsi) are unclear. To address these challenges, our primary goal is to develop state-of-the-art (sota) AI models that can accurately identify pathogenic variants affecting DNA repair genes and predict cancer risks over the life course of high-risk carriers, thereby optimizing screening and prevention strategies in an elsi-informed manner. Our multidisciplinary team comprises experts in computational genomics, AI/ML, health informatics, statistical genetics, medical genetics, population health, oncology, and elsi research from icahn school of medicine at mount sinai (ismms), boston children's hospital/harvard, and columbia university, and has complementary and extensive experience in consortium and team science projects. In our proposed project for magen, aim 1 will develop robust genomic AI models for identifying protein-disrupting missense variants that confer high cancer risks. Aim 2 will combine other genetic factors, including common and rare variant polygenic risk scores (PRS), and non-genetic factors, including EHR, longitudinal lab markers, sdoh, and digital pathology, to predict cancer risk over the life course and optimize screening recommendations for carriers. Aim 3 will cross-validate AI models in real-world population biobanks and determine their clinical impact. Aim 4 will construct an elsi framework and conduct elsi projects to evaluate the multi-faceted impacts of AI-driven genetic diagnostics. Aim 5 will disseminate AI model/predictions, cross-validation data, and elsi recommendations. The completion of these aims will bring genomics-based and multi-modal AI closer to the advancement of personalized medicine in real-world settings by more accurately classifying pathogenic variants, optimizing the timing of screening, and identifying key lifestyle and medical prevention strategies that could ultimately save lives from cancer.