Cancer, cardiovascular disease, diabetes (types 1 and 2), neurological disorders, and kidney disease present vast health and economic challenges. It is thus important to diagnose diseases early and tailor treatments precisely to the diverse needs of patients. In addition, no major disease exists in isolation, motivating a holistic understanding of the biology of each person.
We seek to use UK Biobank data to (i) identify molecular and physiological biomarkers predictive of disease predisposition, onset, and progression, and (ii) uncover cross-disease biosignatures that reveal shared biological connections.
Under Dr. Dimitris Bertsimas, our group has a proven track record in health AI. We demonstrated the strengths of holistic machine learning (ML) approaches, (i) developing a ML framework that integrates multimodal data (e.g. labs, notes, images) to reliably predict patient outcomes (Integrated Multimodal Artificial Intelligence Framework for Healthcare Applications, Soenksen et al., 2022), (ii) extending this to multitask across multiple clinical endpoints (M3H: Multimodal Multitask Machine Learning for Healthcare, Bertsimas & Ma, 2023), and (iii) writing an ML textbook on improving healthcare delivery (The Analytics Edge in Healthcare, Bertsimas, Orfanoudaki, & Wiberg, 2023).
We believe we can incorporate cellular-level data with outpatient visit data to create a more comprehensive and holistic biological picture of each patient. We aim to identify biosignatures for early, accurate, and minimally invasive disease detection, by integrating multiomics data and clinical phenotypes to identify clinical-molecular associations that can better model disease trajectories.
Through this unified ML approach, we aim to build interpretable data-driven models that inform early detection and personalized treatment strategies. Our expertise and experience uniquely position us to uncover both shared and distinct etiologies across these critical disease categories.