Summary and discussion of: “Capturing Heterogeneity in Gene Expression Studies by Surrogate Variable Analysis”

Gene expression study is well known to focus on finding association between expression levels of particular genes and some interesting variables, for example, a disease state. In such studies, besides the primary variable of interest, some other covariates are usually measured and included in the model of association tests. However, it is not possible to measure all the variables related to gene expression. In addition to the known and measured variables, there might be some other unknown and unmeasured factors, which would contribute to the heterogeneity of expression levels of particular genes. Their effect cannot be simply ignored as the unmodeled factors, either biological or technical, may induce extra variability in the gene expression, which would decrease the power to detect the association between gene expression and the primary variable. Or they might introduce spurious signals of the association as the variation on the gene expression could be confounded by a unmodeled factor and the primary variable.The model of Surrogate variable analysis(SVA) was designed to address this problem. In this paper, SVA was shown to able to identify and estimate the components of Expression Heterogeneity(EH), and parse the signal and noise more accurately and reproducibly.