AM25-SN-27-L - Revolutionizing Donor Selection: Integrating AI, Omics Data, and Large-Scale Cohorts for Improved Cell Therapy Outcomes

Identifying donors who can consistently provide high-quality cellular starting material remains a major challenge in cell therapy. Progress is limited by gaps in molecular profiling and fragmented data sharing. Without large datasets, it’s difficult to define the critical donor attributes (CDAs) that predict high-quality starting material.

Current donor selection practices are rooted in isolated success stories, as few industry-wide studies exist. In fields like monoclonal antibody development, computational modeling has helped identify predictive features. While donor selection for cell therapy is a more complex, today’s AI-based tools are also far more capable.

Artificial Intelligence (AI) mimics human intelligence to solve problems while Machine Learning (ML) identifies patterns from large datasets and improves predictions over time. In this context, AI/ML can detect complex relationships between donor traits and therapy outcomes that traditional methods often miss.

Because apheresis is a complex, time-intensive, and costly, screening should focus on easier to obtain samples such as blood, urine, and demographics. Further, given the diversity of cell therapies, it is crucial to identify biomarkers that align with the desired cell subpopulations (e.g. gamma delta T cells) to match specific program goals.

AI/ML can transform donor selection by identifying biomarkers that predict manufacturability and quality of cellular starting material. Recent advances allow models to reduce complexity of an analysis while retaining key information and avoiding overfitting. However, for AI models to generate meaningful insights, comprehensive, well-annotated, standardized donor data set are essential.

This session will explore strategies to build such datasets, apply AI/ML to donor profiling, and enhance manufacturing and patient access. We’ll define CDAs, examine data sources (e.g., registries, blood banks), discuss privacy-respecting models for data use, and explore real-world examples from transplant and regenerative medicine.