CAREER: Integrating Heterogenous Health Data for Improved Predictive and Explainable Methods

About This Grant

Understanding and improving human health is a complex challenge because it involves multiple types of information, including medical records, images or scans, laboratory tests and genetic data. Each type of information provides a distinct piece of the puzzle. However, these pieces don’t always fit together easily because they come in different forms and different quantities and time scales. These differences make it challenging for doctors and researchers to obtain a comprehensive understanding of a person’s health and determine the most effective treatment. This project aims to develop new computational methods that can combine all these types of health information to better predict diseases and design effective treatments tailored to each individual. By improving how these diverse health data can be used, this research could lead to earlier diagnosis, more personalized care, and ultimately better health outcomes for patients. Additionally, the project will involve students in this work to teach them how to use these advanced tools, helping to build a future workforce capable of creating the technology that tackles complex health challenges. This project addresses two major challenges for developing integrative machine learning for health applications: effectively modeling the complex relationships within and between different data types and addressing the sample size imbalances commonly found in real-world datasets. The project approach involves building graph-based frameworks to integrate gene-gene interaction networks into counterfactual explanation methods, enabling precise identification of key genes for therapeutic targeting. Simultaneously, the investigator will embed knowledge of drug-drug interactions into large language models to enhance the prediction of adverse effects and guide treatment optimization. To address the heterogeneity and imbalance across modalities, such as imaging, clinical notes, and genetic screenings, the investigator will design novel joint representation learning techniques. The investigator will also evaluate explainability strategies tailored to multimodal models to improve the interpretability of predictions. These methods will be validated across diverse health datasets and tasks. This research will be closely linked with interdisciplinary educational initiatives, integrating novel multimodal approaches into student training and outreach programs, thereby fostering a synergy between research innovation and workforce development. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.