Fig. 5

Impact of various dataset-related experimental factors on performance of workflows. Each heatmap shows the number of times (cell color) each workflow (row) outperforms other workflows as a particular experimental factor pertaining to the input datasets is varied (columns), when the resulting coexpression networks are evaluated based on the tissue-naive gold standard. The darkest colors indicate workflows that are significantly better than the most other workflows. In addition, the top 5 workflows in each column are marked with their rank, with ties given minimum rank. The heatmaps on the top (a–d) correspond to datasets from GTEx resampling and those on the bottom (e–h) correspond to SRA datasets. The heatmaps from left to right show workflow performance by sample size (a, e; number of samples used to make the coexpression network), sample similarity (b, f; median spearman correlation of 50% most variable genes between samples), read count diversity by counts (c, f; standard deviation of counts sums across samples), and tissue of origin (d, h). Figure S7 contains these heatmaps based on the tissue-aware gold standard