DESeq2 library composition adjustment in size factors
1
0
Entering edit mode
twfs • 0
@twfs-24066
Last seen 2.3 years ago
United Kingdom

Hi folks,

The DESeq2 vst rlog size factors computations account for differences in library composition amongst other things. Library composition differences between sample groups are one of the reasons for not using TPMs. However, we are often presented only with TPM values. I wondered if there are existing methods for quantifying a library composition problem from a set of TPM values and if not what might be a good way to quantify the risk of using the TPMs (for visualisation or other purposes). For example, would starting off by computing a matrix of pairwise Kolmogorov-Smirnov tests for sample count distributions and find significant differences do the trick?

Many thanks

Tim

deseq2 normalization • 1.2k views
ADD COMMENT
0
Entering edit mode
@mikelove
Last seen 14 hours ago
United States

Do you only have TPM and no indication of the sequencing depth?

In tximport, one of the options is to scale up TPMs to account for the sequencing depth (e.g. number of reads aligning to the transcriptome). You can consult Soneson et al 2015 for the details, it performs as well as what we have in the DESeq2 vignette for tximport use, which is counts + offset.

ADD COMMENT
0
Entering edit mode

Thanks Mike.

Yes this would be for where you have just TPMs available and want to check to see if library composition is a problem or not.

But wrt tximport - are you saying that tximport->TPM alone will account for a library composition effect in an equivalent way that the median of ratios does? So if you know your TPMs were calculated that way then you're on safer ground?

ADD REPLY
0
Entering edit mode

No. That is just how the data is passed to the methods which then do their own normalization/offset approach. If you do TPM + library size information -> tximport scaledTPM approach -> VST or rlog, then the last step will perform appropriate median ratio scaling and transformation that stabilizes variance.

ADD REPLY
0
Entering edit mode

Ah ok that makes sense. Normally wouldn't have library size information and only TPMs and so want to try and quantify likely issues, e.g. have an application that already uses TPMs and want to add a quick test to quantify risk by detecting potential library composition effect. Perhaps the K-S test will suffice? Thanks for helping.

ADD REPLY
1
Entering edit mode

Hmm, not sure about a test. Good luck!

ADD REPLY

Login before adding your answer.

Traffic: 702 users visited in the last hour
Help About
FAQ
Access RSS
API
Stats

Use of this site constitutes acceptance of our User Agreement and Privacy Policy.

Powered by the version 2.3.6