Fig. 3
Expression data and four different types of batch correction. From top left to bottom right: PCA Abundance, shows the uncorrected PCA of the rlog normalized counts quantified by salmon and imported to a deseq2 object, next to it in the top right panel a bar plot shows the Low-Quality probability Plow for each sample. The PCA corrected with batch uses the AC-PCA package [21] to return principal components that were computed with the true batch as a confounding factor. The PCA corrected with Plow uses the AC-PCA package likewise, but Plow as a confounding factor. To the right of either corrected PCA we see a corrected PCA on the basis of the data without outliers. For the correction with the real batch, we removed outliers identified from either the base PCA or the corrected version. In Plow we also removed outliers based on the corrected PCA, but additionally added a threshold for Plow after manual inspection of the bar plot on the top right. The last two panels are the PCA corrected by both batch and Plow and the Boxplot of Batch against Plow