In addition to the microarray-based coexpression, we newly prepared RNAseq-based coexpression data, which have high potential, especially for genes with low expression levels. The construction of the RNAseq-based coexpression data has been slightly modified from our previous method applied to Arabidopsis (8). We first selected the RNAseq data entries using the SRAdb R/Bioconductor package (21,22) with the following options: platform = ILLUMINA, library_strategy = RNA-Seq or TRANSCRIPTOMIC. The selected RNAseq entries were downloaded from the DDBJ Sequence Read Archive (23). Then, the FASTQ data were mapped onto the NCBI RefSeq mRNA sequences (24), using Bowtie2 (25). To cover the genes with lower expression level, runs including large number of reads were selected (total mapped counts >10 000 000), resulting 5626, 3746 and 754 runs for human, mouse and fly, respectively. The mapped counts were summed for each gene model and used as the gene expression value. Genes with lower levels of expression; i.e. with average counts across all runs <30, were omitted. After conversion to a base-2 logarithm with a pseudo-count of 1, quantile normalization was applied to the data
