CISE-ANR: IIS: Small: Federated Learning for AI-driven Rare Disease Recognition for Ciliopathies

Rare diseases are individually rare, yet collectively affect millions of individuals worldwide. They pose persistent diagnostic challenges due to variability in clinical presentations, limited clinical expertise outside specialized centers, and fragmented patient data. These challenges are further compounded by small and geographically dispersed patient cohorts and strict privacy and regulatory constraints that limit data sharing across institutions and national borders. This makes it difficult to develop robust and widely applicable diagnostic models at any given site. This project addresses these fundamental barriers by developing a privacy-preserving artificial intelligence framework for identification of rare genetic disorders from observation alone, using a genetic disease of cell component, cilia, as a representative and scientifically challenging disease class. The central goal is to enable large-scale learning from distributed clinical data without direct exchange of patient-level information, reducing diagnostic delays and supporting faster and more accurate clinical decision-making. The proposed research develops a federated learning architecture that enables collaborative model training across multiple institutions while ensuring that sensitive patient data remain local and secure. A core technical contribution is deep clinical phenotyping through accurate extraction, normalization, and representation of observational information from both structured and unstructured electronic health records, leveraging multilingual natural language processing and large language models. These data representations are integrated with biomedical ontologies and rare disease knowledge bases and combined with patient similarity modeling to support rare disease recognition across a wide range of healthcare environments. The project includes algorithm development, cross-site federated evaluation, and real-world validation in collaboration with health care experts. The resulting methods and open resources are designed to generalize beyond the initial single group of diseases to a wide range of rare diseases. It will advance fundamental research in privacy-preserving machine learning and trustworthy AI while contributing to improved population health, international scientific collaboration, and responsible deployment of artificial intelligence in healthcare. 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.