Fully Loaded

COVID-19 Community Update

For the latest updates & information visit Duke’s Coronavirus Response website.

News & Events

Now in Print: Method for High Dimensional Gene Expression Data

Now in Print is a series highlighting recent publications by StatGen’s members.  If you want to learn more about a particular article or researcher, please contact us at statgen@duke.edu.

A new statistical method developed by StatGen member, Jichun Xie, PhD, uses conditional quantile associations to capture a wide range of general association patterns by using a new test statistic called the sample quantile contingency (SQUAC) statistic.  Xie outlines how this method allows for computationally efficient multiple testing, false discovery rate (FDR) control and serves to further statistical capabilities related to inference on high dimensional association networks in J. R. Statist. Soc. B article, “False discovery rate control for high dimensional networks of quantile associations conditioning on covariates.”

What does this method accomplish?

 [JX]:  Past methods for high dimensional association networks allowed for modeling of general associations, but were limited by not allowing for adjustments of covariates.  These methods were further limited by complex general association estimators, which resulted in challenging asymptotic null distributions.  Overall, these methods presented drawbacks in controlling the type I error for general association inferences.

Why is it significant for the field?

[JX]:  This method serves to overcome common challenges associated with gene coexpression network analysis, and specifically challenges with high dimensional gene expression data.  It provides a computationally efficient multiple-testing procedure to infer high dimensional sparse general association network conditioning on covariates.  The methods can be used to study non-Gaussian non-rank-associated data, such as RNA sequencing data. Ultimately, it efficiently provides a solution to address common, but limiting, challenges associated with this type of analyses.

What are the next steps to further explore this topic?

[JX]:  Next steps include making this method more adaptive so that even without specifying quantile points, it can automatically identify global or local conditional quantile associations. We will also investigate new methods to identify gene coexpression network differences of two subpopulations.


Dr. Xie joined the Department of Biostatistics & Bioinformatics as an Assistant Professor in July 2014 and has been involved with StatGen since its inception.  Her research extends beyond StatGen with an affiliation with the Duke Cancer Institute and collaborations with numerous genomics and cancer researchers.  Dr. Xie also teaches several PhD and MB courses in the Biostatistics & Bioinformatics department and serves as a mentor and committee member for students in both programs.

To learn more about Dr. Xie’s research or how to collaborate, please contact StatGen at statgen@duke.edu



Xie, J and Li R. “False discovery rate control for high dimensional networks of quantile associations conditioning on covariates.” J. R. Statist. Soc. B 19 July 2018 https://doi.org/10.1111/rssb.12288.