How to detect mismatch before mapping in RNA-Seq data

Question

In the methods of this paper, the authors say:

Reads were then filtered based on quality score in the UMI region. Any read with >1 low-quality base (phred <=10) were discarded. Reads with more than one mismatch in any of the three 8 nt cell barcodes were also discarded. The cDNA reads (Read 1) were then mapped to either a combined mm10- hg19 genome or the mm10 genome using STAR (44).

I understand how to filter based on mismatch after mapping but in this case it is done before mapping. Has this to be interpreted as "Reads with low quality in more than one base in any of the three 8 nt cell barcodes were also discarded"?

Answer 1

At each step in the SLiTing and pooling the barcodes added are not random, but are taken from a predefined list of possible barcodes.

When the reads arrive, first you filter by reads that have >1 low quality bases in the UMI region (which is truely random).

Then you compare the sequences in the 3 cell-barcode regions against the list of allowed barcodes for each of those positions. If a read is not within an edit distance of one from one of those barcodes, it is discarded.

Finally the reads that pass these filters can be aligned.

Answer 2

It should not be interpreted as "Reads with low quality in more than one base in any of the three 8 nt cell barcodes were also discarded". They only care if the cell barcodes match what are valid cell barcodes within 1 mismatch, the individual qualities are ignored. This sort of filtering turns out to be pretty common in practice, since you have to write a maximum edit distance into whatever you use to demultiplex by cells (or labeling each read, as was likely done in this case).

Answer 3

Cell barcodes =/= sequence reads. The default settings on Illumina's bcl2fastq work exactly the same way; they allow a read to count for a sample if there is a single mismatch in the index, but not more.