Hello,
I have some fastq files produced by NextSeq. The average fragment length is between 350-400bp. The library prep adds 139bp of adapters, so the inserts are 200-250bp.
Would the kallisto command look like this?
kallisto quant -i index -o output --single -l 350 -s 50 R1.fastq.gz --rf-stranded
Thank you
•
written
by
bioinfo •
I made a comment on this post, but that turns out to be not quite right.
I was told via other sources that Kallisto will apply a correction based on the fragment length, basically ignoring alignments that appear to not "fit" into the transcript when the whole fragment is considered.
If so, the value should reflect the original fragment length corresponding to the biologically relevant template and not the artificial construct with ligated adapters.
•
written
by
Istvan Albert
TPM normalized counts (which is what kallisto outputs) requires fragment length information. This is because, in TPM, counts are divided by the "effective length" rather than the actual transcript length. The effective length is the number of positions a fragment can start along a given transcript.
This is the primary reason kallisto requires fragment length information (for paired-end data, fragment length information is inferred automatically from your reads).
I always thought the effective length is computed based on read lengths (subtracting half the read length from each end).
As you and others pointed out, there is a second effective size correction on the 3' end based on fragment size -
I think it is a correction that could have a more substantial effect in some circumstances, for example, when the transcripts are incompletely characterized - which is very common for many organisms: we know the coding regions but not the full transcripts. In those cases, it could be better to "lie" about the fragment size and claim it to be either shorter or have a larger standard deviation - that way it would allow all the data to be used.
No. You, the user, supply the fragment length distribution numbers; and trimming adapters will not affect what position in a reference transcript a read gets mapped to. For a given -l and -s, you'll likely end up with the same TPMs with and without adapter trimming.
Thank you so much sdull! I'm still just a little confused on why you're recommending the insert size parameters (-l 225 -s 25) over the fragment size parameters (-l 350 -s 50) if it doesn't matter if the adapters are included or not. (I've read your comments like 10 times and it's just not sticking sorry)
Because reads can only originate from the insert, not the adapters.
If you have an adapter that's a million base pairs long, you're not going to set -l to one million right? (when the reads can only only originate from a few hundred bp's within the fragment).
(P.S. All that being said, if I'm to be honest, these fragment models [which originated from the cufflinks days] are very poor models anyway so honestly just be semi-arbitrary with it -- it probably won't make much a difference. I might write a paper on their shortcomings at some point.)
Additional topics to the "shortcomings" paper:
Log in to answer this question.
my hunch is that the fragment length only matters when the run is in paired-end mode - in single-end mode I don't see how that would have any effect whatsoever
The kallisto manual says that you need to specify both -l and -s for single end mode. I am just not sure if the values I chose are correct.