TODAY!!! Department of Preventive Medicine Biostatistics Seminar Series: Causal Models for High-Dimensional Omics Data

The Division of Biostatistics of the Department of Preventive Medicine, UTHSC, invites you to attend TODAY's seminar.

Time: Monday, February 23, 2026, 2:00PM - 3:00PM CT

ZOOM Virtual Room Connection: Register in advance for this meeting to get the Zoom Link

Seminar Website: https://www.uthsc.edu/preventive-medicine/events.php

Speaker Bio: https://directory.sph.umn.edu/bio/sph-a-z/tianzhong-yang

Causal Models for High-Dimensional Omics Data

Tianzhong Yang, PhD

University of Minnesota

This talk presents two statistical frameworks for elucidating disease mechanisms using high-dimensional omics data. In the first part, I will introduce pseudotime-dependent transcriptome-wide association studies (pt-TWAS), which integrate GWAS data with single-cell gene expression to identify genes and cell stages underlying disease risk. Pt-TWAS models gene expression as a continuous function of pseudotime and leverages shared genetic effects across cell stages. We show that pt-TWAS captures dynamic genetic effects along developmental trajectories and achieves higher statistical power than pseudobulk TWAS analysis where single-cell gene expression is averaged within each cell types. Applied to B-cell acute lymphoblastic leukemia, pt-TWAS replicates known risk genes and pinpoints their relevant developmental stages. In the second part, I will present a new total mediation effect measure for high-dimensional omics mediators in case-control studies. Existing high-dimensional mediation methods largely focus on continuous outcomes, with limited applicability to binary outcomes and case-control designs. Moreover, many rely on sparsity assumptions that are ill-suited for settings with numerous weak mediators and small, non-sparse effects. We propose a variance-based total mediation effect measure within a liability framework that avoids cancellation of opposing effects and is invariant to disease prevalence. To enable valid inference in case-control studies, we develop a cross-fitted, modified Haseman-Elston regression-based estimator that yields consistent estimation in the presence of non-mediators and weak effects, without requiring exact mediator selection. In an application to the Women's Health Initiative, we find widespread weak mediation of the BMI-coronary heart disease association through metabolomics, with an estimated 89% of the explained liability mediated by measured metabolites.