Abstract:Mendelian randomization (MR) is widely used to infer causal effects of exposures on complex diseases. Increasingly, MR has been applied to molecular traits, such as gene expression, DNA methylation, splicing, and alternative polyadenylation, to identify putative disease genes and regulatory mechanisms. However, transcriptome-wide MR presents several challenges: cis-QTL instruments are often limited, horizontal pleiotropy is widespread, and emerging resources such as developmental transcriptomic datasets frequently have small sample sizes.
In this talk, I will present two related MR frameworks designed to improve causal inference for molecular traits. First, I will introduce FusioMR, a robust Bayesian MR framework applicable to both molecular and complex trait exposures. The single-outcome model, FusioMRs, incorporates gene-region-specific empirical priors informed by QTL strength, linkage disequilibrium, and consistency of genetic effects, enabling robust inference when instruments are sparse. The multi-outcome model, FusioMRm, jointly analyzes correlated diseases, subtypes, or comorbid outcomes, leveraging shared instruments and correlated pleiotropic effects to improve invalid instrument detection and enhance power, particularly for underpowered outcomes. Applications of FusioMR identify cell-type-specific gene expression traits associated with Alzheimer’s disease, alternative polyadenylation events affecting atrial fibrillation and ischemic stroke, and lipid effects on ischemic stroke across ancestries.
I will then discuss dynamicMR, a framework for identifying age-varying causal gene effects by integrating developmental GTEx, adult GTEx, and GWAS summary statistics. By borrowing information across developmental stages and tissues and incorporating gene-embedding-informed empirical priors, dynamicMR improves power under small-sample settings and distinguishes development-specific from adult-specific disease effects. Applications to asthma, pneumonia, and type 2 diabetes reveal distinct temporal patterns of genetic risk, highlighting the importance of developmental gene regulation in complex disease etiology.
Bio：Lin Chen, PhD, is a tenured Professor of Biostatistics in the Department of Public Health Sciences at the University of Chicago. Her research develops statistical and computational methods for integrative genomics, causal inference, and multi-omics data analysis, with a goal of uncovering molecular mechanisms underlying complex traits and diseases. Dr. Chen received her PhD in Biostatistics from the University of Washington and completed postdoctoral training at the Fred Hutchinson Cancer Research Center. Over the past decade, Dr. Chen has worked broadly on multi-omics multi-context integrative methods, causal inference methods for genomic data, genetic regulation across tissues, gene-environment interaction, missing-data methods for omics studies, and integrative proteogenomic and epigenomic analyses. Her recent work includes robust and integrative Mendelian Randomization frameworks for identifying causal molecular effects across tissues, cell types, age groups, and disease outcomes. She is the PI of NIH-funded projects on integrative multivariate genomic analysis and methods for the developmental Genotype Tissue Expression project.
