Detect transcript isoform abundance for a specific gene in scRNA-seq

Question

I want to detect the count of isoform transcripts for a specific gene in scRNA-seq data. Data is coming from cells of Mus Musculus. For transcript isoforms I mean the different alternatives provided by known annotation. For each one I need the total transcript count in each cell.

By looking at the count matrix, alternative transcripts are not defined (see for instance this dataset).

Should I come back to the fastq? Which tool may I use to produce a count matrix with counts per each alternative transcript?

Answer 1

Transcript quantification is a difficult enough problem as it is, when you add the extra difficulty of going from the low read numbers available in scRNAseq if gets even more difficult.

Added to this, many scRNAseq techniques (including 10X, which is what the linked dataset is) only capture the 3' end of the gene, so you can distinguish transcript with different 3' end, but the data simply doesn't exist for doing much more with standard tools.

Lior Pachter does claims that there is benefit in looking at "transcript compatibility counts" rather than just gene counts, and that you can use his Kallisto tool for this, and the these 3' end tagging tools do give you some data for the rest of the transcript.

You could also isolate transcripts with differing three 3' ends and map reads to these using the something like the UMI-Tools pipeline.

Alternatively you could look at things on a junction level - you won't capture all junctions, but you may be able to look at differential junction usage for those you do capture using something like SUPPA or MISO. For this I would use UMI-Tools to get de-duplicated BAM files, split them on cell identity and then pass them into your tool of choice. I don't really how much you will capture.

The last thing to say is that this might be a space worth watching for future developments (cough).

Data from some other single cell methods, like SMART-seq2 or C1 will sequence the whole transcript and you may be able to apply standard tools such as Salmon, Kalisto or RSEM to get transcript usage for them, although as I said, transcript level quantification is noisey, single cell data is noisey, single-cell level transcript data will be very noisy.

Whatever, you are going to have to go back to the raw sequencing data.

Disclosure: I am the author of UMI-Tools.

Answer 2

Looking for alternative splicing in scRNA-seq is difficult, but there is at least one tool that is specifically designed for it: DISCO. The site does not have too much info, but there is a presentation about it. It uses MISO output as the input, which in turn uses SAM.