Fig. 2

Reference Standards demonstrates high concordance between some techniques and major inconsistencies among others. Each method’s pairwise concordance with one another (identity overlap of DEGs) is shown, with the diagonal entries as the total number of DEGs of each respective method, demonstrating the vulnerability of studies relying on a single method to develop a reference standard. The pairwise intersections were calculated using the count of DEGs in the methods of each column as the denominator. The heatmap is approximately symmetric given the different denominators of comparing edgeR’s intersection with NOISeq vs. comparing NOISeq’s intersection with edgeR. In both Yeast (n = 30) and MCF7 (n = 4), edgeR, NOIseq, and DESeq show the best concordance to one another, while DESeq2 has the least concordance to any other method. DESeq2 shows the lack of agreement between what it considers DEGs and the rest of the methods, whereas in the left panel, both DESeq2 and DEGseq differentiate themselves from the cohort. This highlights the need for a consensus as some methods might make certain DEG calls that other methods miss and vice-versa. A conservative approach would be the intersection of all whereas an anti-conservative approach would take the union