Unravelling genetic basis of comorbidity using EHR-linked biobank data

The National Institute of General Medical Sciences awarded $2,373,884 to the University of Pennsylvania on September 1, 2026, under the Biomedical Research and Research Training program (CFDA 93.859) to develop a scalable, multimodal framework for modeling and predicting complex disease interactions using electronic health records and biobank data. The research program constructs disease-disease networks—graph-based models capturing shared genetic, molecular, and clinical relationships across conditions—to advance precision medicine and address comorbidity, which currently impedes accurate diagnosis, treatment selection, and outcome prediction. The work advances three synergistic pillars: constructing modality-aware disease-disease networks across multiple independent cohorts using multimodal data; translating those networks into individualized risk prediction tools using graph-based machine learning; and incorporating plasma proteomics as a dynamic biomarker layer to support real-time disease activity modeling. Rather than focusing on a single disease, the program creates general-purpose, interpretable tools for risk stratification and personalized prognosis across a wide spectrum of conditions by linking static genetic risk with dynamic molecular signals. The award covers performance through August 31, 2030, with work performed in Philadelphia, Pennsylvania.

Abstract

This research program aims to develop a scalable, multimodal framework for modeling and predicting complex disease interactions to advance precision medicine. Comorbidity-the CO-occurrence of multiple chronic conditions-remains a major barrier to accurate diagnosis, treatment selection, and outcome prediction. To address this, our lab constructs disease-disease networks (ddns): graph-based models that capture shared genetic, molecular, and clinical relationships across diseases. Building on foundational work with electronic health records (ehrs), genomics, and biobank-scale data, this program advances three synergistic pillars: (1) constructing modality-aware ddns across multiple independent cohorts using multimodal data; (2) translating ddns into individualized risk prediction tools using graph-based machine learning; and (3) incorporating plasma proteomics as a dynamic biomarker layer to support real-time disease activity modeling. Rather than focusing on a single disease, this program creates general-purpose, interpretable tools for risk stratification and personalized prognosis across a wide spectrum of conditions. By linking static genetic risk with dynamic molecular signals in a unified, flexible architecture, this program will uncover fundamental principles of disease connectivity and progression, supporting both hypothesis-driven discovery and clinical decision-making.