kopardev
commented
3 years ago Currently, CircExplorer2 is run in two steps:
- Parse: This takes in the 2-pass STAR output (Chimeric.out.junction files) and produces the BSJs in bed format (back_spliced_junction.bed). We are not using the align option of CircExplorer2 as it uses the somewhat obsolete TopHat2/TopHat-Fusion pipeline. FYI, I have recently updated the upstream STAR version to 2.7.6a, which includes the --chimMultimapNmax 10 parameter (an updated chimeric read detection algorithm)
- Annotate: The back_spliced_junction.bed file from parse run from the previous step is then annotated using CircExplorer2’s annotate subcommand. CIRCexplorer2 annotate extracts fusion junction reads exactly matching the boundaries of exons of the same isoform by default. As we are also setting --low-confidence, it will also extract fusion junction reads matching the boundaries of exons of the different isofoms of the same gene, and output them in low_conf_circularRNA_known.txt. I believe these 2 steps are fine if you are looking for human circRNAs, but not for viral circRNAs. As viral circRNAs may result in non-canonical BSJs (i.e. BSJs not starting/ending at annotated exon boundaries), I do expect most viral BSJs to be unannotated and hence missing from the circularRNA_known.txt files. As these files (circularRNA_known.txt) are used to create the aggregated count matrix and its annotated version (circExplorer_count_matrix*.txt files), they will be missing some key results from the BSJs. I think what I need to do is to aggregate over all BSJs reported and not just those filtered by the annotate subcommand. In fact, I may completely do away with the annotate rule or replace it with my own custom annotation script which considers all BSJs.
Taka reported:
I still have a bit of problem understanding count_matrix_with_annotations.txt. Is this combining all BSJs (back-splice junctions) detected, or only “known” BSJs are included? if you look at circExplorer_count_matrix_with_annotations.txt, there are no entry from NC_009333.1 (KSHV). But in the rK_LEC_si_Con_1.back_spliced_junction.bed, for example, I see 155 entries. Why are these not included in the aggregated count_matrix file? I think I saw the same in ccbr983_v3.