Abstract: Single-cell technologies now support two complementary modes of studying gene regulation: population-scale profiling across many individuals, which enables analysis of how gene co-expression varies with biological covariates, and perturbation-based experiments such as Perturb-seq, which enable causal interrogation of regulatory relationships. In this talk, I present statistical methods tailored to each setting. For population-level single-cell data, I develop Fréchet regression on the Bures–Wasserstein manifold to model covariance matrix–valued outcomes and test how subject-specific gene co-expression structure changes with covariates. For Perturb-seq data, I introduce an instrumental-variable framework that leverages genetic perturbations to recover causal gene regulatory graphs while remaining robust to unobserved confounders. Applications include aging-related changes of gene co-expressions in CD4⁺ naïve and central memory T cells from PBMCs and causal network reconstruction from CRISPR interference experiments in K562 cells.
Bio: My methods reserach was mostly motivated by problems in genetics and genoimics. I have worked on a variety problems in statistical genetics and genomics, including methods for family-based genetic linkage and association analysis, methods for admixture mapping, methods for genome-wide association analysis, methods for analysis of microarray time course gene expression data, high dimensional regression analysis for genomic data, methods for copy number variation analysis and methods for analysis of next generation sequence data. I have published both statistical methodological research in top statistics/biostatistics journals (JASA, AOS, AOAS, Biometrika, Biometrics, Biostatistics etc ) and in top genetics journals (AJHG, Plos Genetics, etc) and collaborative research in top scientific journals (Science, NEJM, Nature, Nature Genetics, PNAS, Developmental Cell etc).
