Kallisto percent reads pseudoaligned

Question

[genomerik]

Hello,

I am using Kallisto for transcript expression quantification and I noticed that I am getting a very low percentage of reads that are pseudoaligning. It is my understanding that output below refers to the number of reads that are being associated with transcripts:

[quant] processed 54,011,492 reads, 709,696 reads pseudoaligned

I am using these parameters:

kallisto quant --bias --fr-stranded -i ${index} -o ${opath} ${readline}

Has anyone else seen this problem? Does it have to do with specifying the strandedness of the reads?

Thank you!

[h.mon]

Are you using a proper reference transcriptome? Did you build it yourself, or download from somewhere (NCBI, ENSEMBL)? How were you libraries prepared, which kit? Strandedness will depend on library prep kit.

Yes, it could be due to passing an incorrect strand flag, but it could also be an incorrect reference transcriptome, or it could be sample contamination with foreign RNA, or it could be major rRNA contamination, or it could be you did a DNAseq run and are trying to quantify against transcriptome.

Answer 1

[genomerik]

Ok I figured it out. It was just the simple mistake of using the wrong strand flag. I used "--fr-stranded" in Kallisto because I thought it was the same as "fr-firststrand" from using Tophat2. Actually "--rf-stranded" Kallisto seems to equal "fr-firststrand" Tophat2 in terms of the representation of strandedness in the fastq files.

"--rf-stranded" Kallisto == "fr-firststrand" Tophat2

Next time I should read the manual more carefully. :p

[Michael]

Beware of the reverse strand ;) What is your pseudo alignment rate now?

[genomerik]

This is from a different group of fastq files:

with Kallisto --fr-stranded: [quant] processed 42,461,891 reads, 1,105,842 reads pseudoaligned

with Kallisto and no strand information: [quant] processed 42,461,891 reads, 31,602,302 reads pseudoaligned

with Kallisto --rf-stranded: [quant] processed 42,461,891 reads, 30,780,677 reads pseudoaligned

So it is interesting to see that you get quite a few "wrong" pseudoalignments when you don't have stranded data.