Fig. 4
From: Analysis and correction of compositional bias in sparse sequencing count data
Estimation of compositional correction scales from sparse count data. On the left column, we plot the feature-wise ratio (Λgji) estimates adjusted for sample depth from each feature i in one of the samples from the Adrenal tissue of the rat body map dataset (bulk RNAseq), and on the right column, we plot the same values arising from a sample in the Diarrheal dataset (16S metagenomics). The top and bottom rows correspond to the scales estimated using TMM and DESeq respectively. In the case of bulk RNAseq data, large numbers of individual feature estimates agree on a compositional scale factor. Simple averaging, or some robust averaging would help us obtain the scale factor exactly. A similar robust behavior is observed with all the tissues available in the bodymapRat dataset (considered later in text). On the second column, we plot the feature-wise ratio values from a metagenomic 16S marker gene survey of infant gut microbiota. There is no general agreement among the features on the scale factors, and simple averaging will not work. We note that what we have shown are fairly good cases. Several samples entertain only a few tens of shared species with an arbitrary reference sample within the dataset. In this work, we aimed to model this variability and estimate the scale factors robustly by borrowing information across features and samples