An independent component analysis confounding factor correction framework for identifying broad impact expression quantitative trait loci
Fig 1
ICA is used to decompose the gene expression matrix Y into an IC coefficient matrix A and a component matrix S. Associations between the genotypes and coefficients in matrix A are tested to label any candidate genetic effects to be removed from the correction. In the example above, the first IC, shown in red, is marked as a candidate genetic component and the corresponding columns of A and rows of S are removed. Using the lower rank A* and S*, expression values originating from non-genetic components are reconstructed in Y*. Finally, K is created by calculating the sample covariance matrix of Y*, and included as a random effect in the mixed model for eQTL analysis.