Evaluation of critical data processing steps for reliable prediction of gene co-expression from large collections of RNA-seq data
Fig 3
Importance of sample numbers and normalization approaches.
(A) Sample count vs quality of co-expression networks. The quality of individual networks of each cell type and tissue generated by using different workflows are indicated by small points (forming a vertical pattern). Larger points are averages for each dataset. Blue: Mouse, red: Human datasets. (B) Boxplots of the quality of networks made using each of the six normalization methods, in function of dataset size. Datasets were divided into three sets of 36 datasets according to size. Red: Small datasets (20 to 44 samples); Green: Medium-sized datasets (45 to 111 samples); Blue: Large datasets (113 to 2,644 samples).