Many of the most widely-used approaches for relating genomic measurements to disease outcomes involve reducing the genomic data through additive summaries. While additive summaries are unlikely to capture the information in genomic data at its most fundamental level, there are compelling statistical arguments supporting the use of such approaches in practice. I will discuss some aspects of summarizing gene expression measurements using both additive and non-additive signatures in the context of our work to identify predictive signatures of disease progression in chronic kidney disease (CKD).
First we demonstrate the existence of linear functions of gene expression that predict renal performance, and provide some results that indicate how these correlations may improve when larger data sets become available. Second, we consider the marginal correlations between the expression of individual genes and disease outcomes, and demonstrate that many genes have different marginal correlations in the different CKD subtypes. Third, we consider complementary information in multiple genes, and assess how information from several genes may combine to improve a prediction. Time permitting, we will also discuss a particular type of non-additive relationship that can potentially be estimated even in modest data sets.
This was presented at the ISN Forefronts Symposium event “Systems Biology and the Kidney” that took place from 7-10 June 2012 in AnnArbor , Michigan, US.
Additional Info
-
Contains Audio:
Yes -
Source:
ISN -
Event:
Forefronts -
Year:
2012 -
Members Only:
No