Adapt to challenge of scRNA-seq data mostly consisting of single-end reads

[haessar]

A lot of our assumptions have relied on using paired-end reads. It turns out that the Cell Ranger BAM outputs from #2 are actually single-reads, which are incompatible with the pipeline developed for #9.

There are three options:

1. Apparently long-read single-cell data exists, which stringtie appears to have compatibility with. If we could get our hands on some, we could adapt the pipeline from (WP6) Explore isoform clustering tools #9 to run stringtie in long-read mode, and add any other alignment component necessary (e.g. minimap?)
2. Obtain some bulk RNA-seq data that is already tissue-specific (i.e. corresponds to different tissue types within the worm). The CB stuff (i.e. (WP2) Scripting to group reads from BAM files by Cell Barcode (CB) #4) would cease to be relevant because those barcodes wouldn't exist anymore, but we would already have distinct BAM files to use in other parts of the workflow to represent different "cell-types".
3. Carry on as we are with the BAM files from (WP1) Use Cell Ranger to align reads from published dataset #2 and just drop parts of the workflow that rely on CB.

We can use this issue to discuss and find a way forward.

[dariober]

It turns out that the Cell Ranger BAM outputs from #2 are actually single-reads, which are incompatible with the pipeline developed for #9

A little more detail on this: We can use single-end reads, but the problem is that scRNAseq captures the 3'end only so these are not very informative for splicing analysis.

[mberriman]

yes there's a bias to the 3' end but we have been surprised that in many cases coverage is along the length. For a given splice junction, there's often a lot of alignable evidence. The lack of read pairs does however mean that phasing full-length transcripts is not possible