Mediation analysis has been a popular framework for elucidating the mediating mechanism of the exposure effect on the outcome. Previous literature in causal mediation primarily focused on the classical settings with univariate exposure and univariate mediator, with recent growing interests in high dimensional mediator. In this paper, we study the mediation model with high dimensional exposure and high dimensional mediator, and introduce two procedures for mediator selection, MedFix and MedMix. MedFix is our new application of adaptive lasso with one additional tuning parameter. MedMix is a novel mediation model based on high dimensional linear mixed model, for which we also develop a new variable selection algorithm. Our study is motivated by the causal gene identification problem, where causal genes are defined as the genes that mediate the genetic effect. For this problem, the genetic variants are the high dimensional exposure, the gene expressions the high dimensional mediator, and the phenotype of interest the outcome. We evaluate the proposed methods using a mouse f2 dataset for diabetes study, and extensive real data driven simulations. We show that the mixed model based approach leads to higher accuracy in mediator selection and mediation effect size estimation, and is more reproducible across independent measurements of the response and more robust against model misspecification. The source R code will be made available on Github https://github.com/QiZhangStat/highMed upon the publication of this paper..

Mapping genotype to phenotype is an essential topic in genetics and genomics research. As the Omics data become increasingly available, 2-variable methods have been widely applied to associate genotype with the phenotype (genome-wide association study), gene expression with the phenotype (transcriptome-wide association study), and genotype with gene expression. However, signals detected by these 2-variable association methods suffer from low mapping resolution or inexplicit causality between genotype and phenotype, making it challenging to interpret and validate the molecular mechanisms of the underlying genomic variations and the candidate genes. Under the context of genetics research, we hypothesized a causal chain from genotype to phenotype partially mediated by intermediate molecular processes, i.e. gene expression. To test this hypothesis, we applied the high-dimensional mediation analysis, a class of causal inference method with an assumed causal chain from the exposure to the mediator to the outcome, and implemented it with a maize association panel (N = 280 lines). Using 40 publicly available agronomy traits, 66 newly generated metabolite traits, and published RNA-seq data from 7 different tissues, our empirical study detected 736 unique mediating genes. Noticeably, 83/736 (11%) genes were identified in mediating more than 1 trait, suggesting the prevalence of pleiotropic mediating effects. We demonstrated that several identified mediating genes are consistent with their known functions. In addition, our results provided explicit hypotheses for functional validation and suggested that the mediation analysis is a powerful tool to integrate Omics data to connect genotype to phenotype.

Mouse knockouts of Cntnap2 exhibit altered neurodevelopmental behavior, deficits in striatal GABAergic signaling and a genome-wide disruption of an environmentally sensitive DNA methylation modification (5-hydroxymethylcytosine, 5hmC) in the orthologs of a significant number of genes implicated in human neurodevelopmental disorders. We tested adult Cntnap2 heterozygous mice (Cntnap2+/-, lacking behavioral or neuropathological abnormalities) subjected to a prenatal stress and found that prenatally stressed Cntnap2+/- female mice showed repetitive behaviors and altered sociability, similar to the homozygote phenotype. Genomic profiling revealed disruptions in hippocampal and striatal 5hmC levels that were correlated to altered transcript levels of genes linked to these phenotypes (e.g., Reln, Dst, Trio, and Epha5). Chromatin-immunoprecipitation coupled with high-throughput sequencing and hippocampal nuclear lysate pull-down data indicated that 5hmC abundance alters the binding of the transcription factor CLOCK in the promoters of these genes (e.g., Palld, Gigyf1, and Fry), providing a mechanistic role for 5hmC in gene regulation. Together, these data support gene by environment hypotheses for the origins of mental illness and provide a means to identify the elusive factors contributing to complex human diseases.

High dimensional mediation analysis has been receiving increasing popularity, largely motivated by the scientific problems in genomics and biomedical imaging. Previous literature has primarily focused on mediator selection for high dimensional mediators. In this paper, we aim at the estimation and inference of overall indirect effect for high dimensional exposures and high dimensional mediators. We propose MedDiC, a novel debiased estimator of the high dimensional overall indirect effect based on difference-in-coefficients approach. We evaluate the proposed method using intensive simulations and find that MedDiC provides valid inference and offers higher power and shorter computing time than the competitors for both low dimensional and high dimensional exposures. We also apply MedDiC to a mouse f2 dataset for diabetes study and a dataset composed of diverse maize inbred lines for flowering time, and show that MedDiC yields more biologically meaningful gene lists, and the results are reproduciable across analyses using different measures of identical biological signal or related phenotype as the outcome. Upon the acceptance of the paper, the code will be available on GitHub (\url{https://github.com/QiZhangStat/MedDiC}).

Hi-C experiments have become very popular for studying the 3D genome structure in recent years. Identification of long-range chromosomal interaction, i.e., peak detection, is crucial for Hi-C data analysis. But it remains a challenging task due to the inherent high dimensionality, sparsity and the over-dispersion of the Hi-C count data matrix. We propose EBHiC, an empirical Bayes approach for peak detection from Hi-C data. The proposed framework provides flexible over-dispersion modeling by explicitly including the “true” interaction intensities as latent variables. To implement the proposed peak identification method (via the empirical Bayes test), we estimate the overall distributions of the observed counts semiparametrically using a Smoothed Expectation Maximization algorithm, and the empirical null based on the zero assumption. We conducted extensive simulations to validate and evaluate the performance of our proposed approach and applied it to real datasets. Our results suggest that EBHiC can identify better peaks in terms of accuracy, biological interpretability, and the consistency across biological replicates. The source code is available on Github (https://github.com/QiZhangStat/EBHiC).