Hi,
I've a bunch of RNA-Seq samples sequenced on a HiSeq2000 (2x100bp) and an other bunch sequenced on a NextSeq500 (2x150bp and 2x76 depending on the samples). The sequencing depth varies between 25M to 60M on average. All libraires were prepared using the same kit (ribo-zero). Only the sequencing kit changes between the HiSeq and the NextSeq.
I applied kallisto on all these samples and now want to check the expression of a specific gene across the samples. Is it correct to use the TPM from kallisto to compare the samples ; or should I use an other metric : or maybe add an additional per-sample normalization step ?
Thanks
•
written
by
Nicolas Rosewick
Can you absorb the effect of sequencer in your design matrix, or does that make it not full rank? TPM is probably the most stable metric to use for a quick eyeball (especially better than R/FPKM). There's the tximport method which allows you to read in Kallisto results and leverage the bootstraps so it's ready for DESeq2
Indeed I could do a PCA plot a see if my samples cluster by sequencer (or not). I'll try it.
Question : I didn't apply bootstrap (-b option in kallisto). Should I ?
I've not tested it's effect in DESeq2 to be honest, but I did notice the other day when looking through the tximport package, that there's a parameter called varReduce, which I think uses the bootstraps. You'll be fine not using bootstraps.
I tested a PCA plot using the TPM and it seems ok to compare samples sequenced on different sequencer (at least in my case).
Log in to answer this question.
