Our group develops statistical methods and software for the analysis of data from high-throughput genomics projects, with particular interest in methods for identifying the genomic mechanisms underlying complex traits. We have considerable expertise in the development and implementation of methods for both static and time course microarray experimental design and analysis, expression quantitative trait loci (eQTL) mapping, gene association network reconstruction, and gene set co-expression analysis. An overview of our more recent work is given below:
RNA-seq is a revolutionary tool for transcriptomics with a number of compelling advantages over microarrays. One in particular is the ability to quantify isoform-specific expression levels. We are currently developing methods that allow for both differentially expressed (DE) genes and isoforms to be identified, with a number of applications focused on studies of breast cancer done in Michael Gould's lab. We are also extending the methods to enable identification of DE patterns over time, in collaboration with Jamie Thomson's lab, where the temporal pattern of expression during various stages of development is of primary interest.
Recognizing that in many cases complex disease may arise from a de- or re-regulation of genes that does not affect average expression levels significantly (or at all), but rather affects the ways in which genes interact within a network, we have developed an interest in methods for identifying differentially regulated networks of genes. One specific type of differential regulation we have considered is differential co-expression (or differential correlation -DC ). We are currently developing an empirical Bayes model based approach to enable the powerful and efficient identification of DC networks; and we are working on extensions to identify genomic regions where allelic variation results in DC between and among transcripts (referred to locally as DC mapping). This has been done in collaboration with Alan Attie's lab.
Statistical Methods for Personalized Genomic Medicine
Technological advances over the past decade allow for unprecedented measurement of genomic and clinical variables that, ideally, may be used to better treat patients. Unfortunately, this ideal has yet to be achieved. A major obstacle is the lack of statistical models that are both predictive of an individual's outcome as well as clinically actionable. Our group is working to develop methods to this end. Much of our work in this area focuses on the cancer genome atlas ovary study in collaboration with gynecological oncologists William Bradley and Janet Rader.