Moreover, variancePartition identifies genes that vary along different aspects of the study design (Fig. 4 c), and visualization of a subset of these genes illustrates the strong expression differences when stratified by sex, cell type and individual (Fig. 4 d–f). variancePartition enables further interpretation of the batch effect because it gives results at a gene-level resolution. The samples were processed in 6 technical batches and this axis of variation explains a median of 29.4% of total variation, indicating a large technical effect. Consistent with other analyses, the fraction of variation explained by batch at the gene-level is positively correlated with GC content (Fig. 4 g).
