Collaborative Research: FDT-BioTech: Mechanistic-Neural Foundations for Digital Twins of Parkinson's Disease Symptom Dynamics Under Levodopa Treatment

Parkinson’s disease affects more than one million Americans and creates major health, caregiving, and economic burdens. Levodopa is the most effective treatment for many symptoms, but its effects become harder to manage as the disease progresses. Patients may cycle between good symptom control, symptom return, and involuntary movements caused by excessive medication. Because clinic visits and symptom diaries cannot capture these changes in everyday life, treatment often relies on trial and error. This project will develop a patient-specific digital twin that integrates established medical knowledge, medication records, and smartphone and wearable data to estimate symptom changes, predict responses to different dosing schedules, and support safer treatment decisions under clinician oversight. The research will reduce medication trial and error, improve the quality of life for people with Parkinson’s disease, and lessen caregiver burden. The project will release software and evaluation tools that can help researchers and regulators assess digital health technologies before clinical use. It will train students in computing, statistics, AI, and health, while engaging the Parkinson’s community to improve usability and trust. These efforts will advance digital health, strengthen the biomedical workforce, and promote national health and welfare. The project will establish foundations for patient-specific digital twins of Parkinson’s disease under delayed medication effects, partially observed clinical states, limited longitudinal data, and safety constraints. First, it will develop a hybrid mechanistic-neural model that links levodopa intake to hidden clinical states and then to gait, voice, touchscreen, and other sensor signals. Second, it will create a posterior-weighted expectation-maximization method to estimate patient-specific response parameters, infer continuous symptom trajectories, share information across patients, and quantify uncertainty through conformal calibration. Third, it will formulate within-day medication planning as a constrained belief-space control problem. A risk-aware planner will evaluate candidate schedules, maximize time in a therapeutic state, limit unsafe-state risk, and abstain or request additional information when uncertainty is too high. The resulting methods and open tools will support continuous monitoring, counterfactual treatment simulation, and safe decision support for Parkinson’s disease and other biomedical digital twins. This project is jointly supported by the National Institutes of Health through the Office of Data Science Strategy. 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.