1
$\begingroup$

I have 20 samples out of which 14 are 100 bp in length and 6 are 150 bp. Is there a way to normalize the read length for cross-sample differential expression comparison? What would be the best way to approach this?

Note: The samples are NOT from reference/model species.

$\endgroup$
1
  • $\begingroup$ Are these paired-end reads (i.e. 2x100bp; 2x150bp), or single-end reads? That will make quite a big difference in terms of the mappability of the reads to transcripts. $\endgroup$
    – gringer
    Mar 9 at 3:12
2
$\begingroup$

The simplest thing to do is to trim the 150 bp fasts so that they are 100 long. I don't think there is an easy way to correct for the fact that the 150 bp long reads will have a higher unambiguous rate of alignment and gene assignment than the 100 bp long reads.

If you have a mix of all the experimental conditions across all the lengths, you can include length as an element in the design, so that DESeq2 may be able to model the difference between the two batches.

$\endgroup$
0
$\begingroup$

I've done this by doing a RPkM-like transformation which I call VSTPk. Transformations that adjust for transcript length will correct a little bit for the noise associated with mapping short reads to transcripts of varying lengths:

https://bioinformatics.stackexchange.com/a/15218/73

Another, probably better, alternative is to use Salmon to map reads to transcripts, which applies its own mapping-based correction to read counts. Salmon produces a statistic that relates to the mappable length of a transcript, and normalises counts based on how many molecules the model predicts came from each transcript isoform.

Unfortunately all these normalisation techniques require good gene models, which are difficult to obtain for non-model organisms.

$\endgroup$
1
  • $\begingroup$ I think this is not what OP is asking. OP has different read lengths, that has nothing to do with gene- or transcript length which RPKM-ish transformations are acting on. The concern here is mappability bias, not the fact that longer transcript inherently produce larger counts. I would therefore go with the suggestion above to simply trim back reads to the same length, that is both simple and effective to eliminate this potential bias without actually applying any data transformation or normalization, plus is independent of gene models and reference annotations. Agree with use of salmon. $\endgroup$
    – ATpoint
    Mar 9 at 11:27

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.