The fraction of expression variation is easily interpretable across genes, drivers of variation and datasets. Thus variancePartition produces a more detailed and quantitative genome-wide overview than the standard principal components analysis (PCA) [10] and hierarchical clustering (HC) [11] approaches. PCA and HC focus on the major axis of variation, and they overlook the secondary drivers of variation that can be well characterized with variancePartition. Moreover, the gene-level results from variancePartition indicate genes that deviate from the genome-wide trend and integration with additional data can enable a further interpretation. While PCA and HC do not give gene-level results, differential expression (DE) analysis reports gene-level fold change and corresponding p-value for each aspect of the study design. Yet DE analysis does not produce a clear genome-wide summary, and the fold change and p-values are not easily comparable across multiple drivers of variation.
