Fig. 1
Exaggerated false DEGs identified by DESeq2 and edgeR from anti-PD-1 therapy RNA-seq datasets. A Barplot showing the average numbers of DEGs (left y-axis) and the proportion of DEGs out of all genes (right y-axis) identified from 1000 permuted datasets. The error bars represent the standard deviations of 1000 permutations. The red dots indicate the numbers of DEGs identified from the original dataset. B The distributions of the number of permuted datasets where a gene was mistakenly identified as a DEG. The percentages corresponding to the numbers are listed in parentheses below the numbers. C Barplot showing the average numbers of DEGs (left y-axis) and the proportion of DEGs out of all genes (right y-axis) identified from both the original dataset and any of the 1000 permuted datasets. The error bars represent the standard deviations of 1000 permutations. The red dots indicate the numbers of DEGs identified from the original dataset. D Percentage of permuted datasets where a DEG identified from the original dataset was also identified as a DEG. The genes are sorted by absolute log2(fold-change) in the original dataset in decreasing order. The absolute log2(fold-change) values corresponding to the ranks are listed in parentheses below the ranks. The line is fitted using the loess method, and the shaded areas represent 95% confidential intervals. E GO term enrichment for the DEGs identified from at least 10% permuted datasets. The top 5 enriched biological processes GO terms are shown. The analyses were performed using R package clusterProfiler. P.adjust represents the adjusted p-value using the Benjamini & Hochberg method. F Violin plots showing the poorness of fitting the negative binomial model to the genes identified by DESeq2 or edgeR as DEGs from ≥ 20% vs. ≤ 0.1% permuted datasets. The poorness of fit for each gene is defined as its negative log10(p-value) from the goodness-of-fit test for the negative binomial distributions estimated by DESeq2 or edgeR. The p-value in each panel was calculated by the Wilcoxon rank-sum test to compare the two groups of genes' poorness-of-fit values