Inflated UMI counts using --umi mode

[leonfodoulian]

Hi,

I am pseudomapping UMI based single-cell RNA-seq data with kallisto. The number of bases per UMI is 4. This should yield a maximum of 256 (i.e. 4^4) counts per transcript. However, when counting UMI per gene and per cell, I am getting inflated UMI counts compared to the maximum expected number of counts (up to 1731 counts, that is 6.7 times more than expected!). I understand that kallisto is pseudomapping onto the transcriptome, and that collapsing transcript counts to gene counts might inflate the maximum expected number of UMI. However, I should assume that it has to internally handle this issue. The command line I am using to run the pseudoalignment is the following:

kallisto pseudo -i Homo_sapiens.GRCh38.cdna.all.idx -o kallisto_output -b batch_file.txt --umi -t 6 2>&1 | tee log.txt

Briefly, to collapse transcript counts to gene counts, I first count the number of transcripts per equivalent class (ec for short), then divide the number of UMI for each ec and for each cell by the number of transcripts in that ec. I then sum the "corrected" UMI counts per gene and per cell. I am using a custom made R script for that. I can share the script if that will help solving this issue. However, I think that the problem (if that is one) lies prior to my processing of the files using the R script.

Anyone has ever encountered this issue? If yes, sharing your experience is very much appreciated.

Thank you in advance!

Best,
Leon