The Department of Statistics and Actuarial Science Spring Colloquium Series presents:

Lynn Lin, Associate Professor of Biostatistics & Bioinformatics, Duke University

"Learning from Multi-Resolution Biomedical Data"

Abstract: Modern scientific studies increasingly collect data with a nested structure, where large collections of fine-scale measurements, such as single-cell profiles, are linked to subject-level covariates and outcomes and, in many settings, are observed repeatedly over time. This combination of cell-level data, subject-level information, and longitudinal structure introduces multiple levels of resolution and poses fundamental challenges for statistical learning, including how to compare heterogeneous samples, how to aggregate fine-scale information for subject-level inference, and how to make reliable predictions in small cohorts.

In this talk, I present methods motivated by single-cell immune profiling studies that address these challenges. First, I introduce an optimal transport-based framework for comparing samples represented as distributions of single-cell measurements, enabling differential analysis that goes beyond summary statistics. Second, I present a predictive modeling approach that learns compact cell-level representations and models their subject-level temporal structure using Gaussian processes, allowing outcome prediction in data-limited settings. Together, these methods address complementary learning tasks arising from multi-resolution data. Applications to clinical immune profiling studies illustrate how explicitly leveraging structure across cells, individuals, and outcomes can improve interpretability and predictive performance. I conclude by discussing challenges in learning from complex biomedical data structures.

Meet and Greet at 3 p.m. in 241 SH. Colloquium at 3:30 p.m. in 61 SH.