Register for: WIMSIG2017

Details of talk

TitleInvestigating non-linear associations in high dimensional ‘omics data
PresenterNicola Armstrong (Murdoch University)
Author(s)Nicola Armstrong
SessionBiostatistics and Bioinformatics
Time13:30:00 2017-09-26
Abstract


In ‘omics studies, the trait or variable of interest is often either binary
(e.g. cancer vs normal), or quantitative (e.g. height). Associations between the
trait and, for instance, gene expression measurements, are then uncovered
through use of either logistic or linear regression. The predominance of these
two methods is partly due to computational issues, including the time required
to run an analysis. However, they also place restrictions on the type of trait
that can be considered: often a categorical classification is collapsed to two
categories only, or a nominal scale is considered to be continuous in order to
enable a fast analysis.  Additionally, for quantitative traits, it may be
limiting to assume that any association is in fact linear. 

Here, we investigate the use of mutual information (MI) to simultaneously
uncover associations both between and within SNP and gene expression data and
report the results. We use data on the Sydney Memory and Ageing Study (MAS)
(n=521). For these participants, SNP genotyping was undertaken using the
Affymetrix SNP 6.0 array and gene expression measured using the Illumina HT-12
array. We focus particularly on several traits that can be coded in different
ways, such as alcohol consumption and smoking status (e.g. never/ever;
current/past; or frequency) and investigate the impact on the associations
identified.