Elucidating the Altered Transcriptional Programs in Breast Cancer using Independent Component Analysis
Figure 1
The ICA Model of Gene Expression
Schematic depiction of the ICA model for gene expression.
(A) Measured gene expression variations are caused by alterations in the activation levels of biological pathways. In the ICA model, the gene expression matrix is decomposed into the product of a “source” matrix S and a “mixing” matrix A, where K is the number of inferred independent components (IC) to which pathways and regulatory modules map. The columns of S describe the activation levels of genes in the various inferred independent components, while the rows of A give the activation levels of the independent components across tumor samples. The product of S and A can be written as a sum over the IC submatrices IC-1,IC-2,...IC-K.
(B) IC–k–submatrix is obtained by multiplying the k-th column of S, Sk, with the k-th row of A, Ak. The genes with the largest absolute weights in Sk are selected and tested for enrichment of biological pathways, while the distribution of weights in Ak are tested for discriminatory power of phenotypes. (Colour codes for heatmaps: red, overexpression; green, underexpression; blue, upregulation; yellow, downregulation.)