Analysis of publicly available gene expression studies demonstrate that variancePartition recovers striking patterns of biological and technical variation that are reproducible across multiple datasets. At a genome-wide level, expression variation across individuals and cell types is large enough to overcome the technical variation of transcriptome profiling. Yet at the gene-level there is substantial deviation from the genome-wide trend due to a range of biological and technical factors. By quantifying the variance attributable to each aspect of the study design, variancePartition facilitates the interpretation of these gene-level effects in the context of additional information. We demonstrate reproducible findings that cross-individual variation is driven by cis-eQTL’s and technical variation across laboratories associated with GC content. Moreover, variation across individuals and the relationship to cis-eQTL’s depend on the cell or tissue type.
