We report a bug in satuRn 1.4.0. (Bioconductor release 3.15). The bug was inadvertently introduced in satuRn 1.3.1 (from the former Bioconductor devel). Note that the bug was not thus present in any of the older Bioconductor releases 3.13 and 3.14 (satuRn 1.0.x, 1.1.x and 1.2.x).
Bug details:
Imagine a gene with three isoforms and two cell types. The goal is to assess DTU between cell types. All isoforms are expressed in all cells of cell type 1. However, none of the isoforms are expressed in any of the cells in cell type 2 (i.e., the gene is not expressed in cell type 2).
satuRn computes the log-odds of picking a certain isoform from the pool of isoforms in each cell type, and then compares these log-odds estimates between the cell types. However, in this example, the log-odds of picking a certain isoform from the pool of isoforms in cell type 2 cannot be computed, as there is no data. Hence, the DTU test statistic should be NA. However, due to erroneous handling of NA estimates, which was inadvertently introduced in satuRn 1.3.1. while aiming to resolve github issue 16, the log-odds in cell type 1 will be compared to zero. Hence, (erroneous) results can be obtained for this contrast, even when there are no data in cell type 2.
Note that in many cases such isoforms may not pass filtering and would not get evaluated altogether. However, when analyzing sprase scRNA-Seq datasets with a lenient filtering criterium, this problem will apply, and will result in mistakes in the inference.
satuRn version for Bioconductor release 3.15
NEW FEATURES
NEWS.md file to track changes to the package.SIGNIFICANT USER-VISIBLE CHANGES
foo() or parameter param.BUG FIXES