Fig. 2.

Quality assessment at the sample level. Multi-dimensional scaling (MDS) plots of the NSC (A) and Lymphoma (B) data sets, with samples numbered and color coded by genotype. Distances correspond to the mean log₂fold-change for the top 500 genes that best discriminate each pair of samples. In both experiments, one or more samples cluster poorly with replicates of the same genotype, motivating the use sample weights (D) in the regression modeling to detect differential expression. Panel C shows a scatterplot of the mean–variance relationship in abundance estimated from biological replicates from the NSC data set using the voom method. Panel D shows the sample weights estimated for the NSC data set that are combined with voom's abundance-related weights in the voomWithQualityWeights function and used in the linear model analysis to detect differentially expressed genes.