Question

How to extract the voom normalized value for each observation?

1

Entering edit mode

harelarik ▴ 60

@harelarik-13564

Last seen 3.2 years ago

Israel

I want to use the output values of voomWithQualityWeights as an input for further analysis (e.g., input for network).

The outputs of voomWithQualityWeights are:

1. The $E should actually be: logCpm+0.5 for each gene regardless of voom normalization.

That does not include the voom normalization.

2. weights- are "numeric matrix of inverse variance weights" .

These are probably only the weights for each observation (observation= gene X sample).

3. $sample.weights- weight normalization per sample

**How do I extract the voom normalized value for each observation? Is it $E*weight of each observation? (just multiplication?)? How should I combine the sample weight?**

Thank you.

voomWithQualityWeights output • 4.6k views

ADD COMMENT • link updated 7.1 years ago by Gordon Smyth 52k • written 7.1 years ago by harelarik ▴ 60

0

Entering edit mode

The 'weights' output from running voomWithQualityWeights() already combines both the observational and sample-specific weights.

These weights are relative though and only make sense in the context of a linear model (i.e. in a per gene analysis).

I would not recommend just multiplying the logCPM values by them and using them in your network analysis.

ADD REPLY • link 7.1 years ago Matthew Ritchie ▴ 1000

0

Entering edit mode

Thank you very much.

I. Are the $E values are "logCpm+0.5" transformation of raw counts or the of TMM values (made by calcNormFactors(y))?

II. Considering one found Differentially expressed genes, based on voom values, and would like to continue to downstream analysis (e.g., network analysis). Which values would you think would be most consistent to use:

* This question was answered by Gordon below.

a. The $E (logCpm+0.5) values?

b. Some other form of variance stabilization values.

For example the out put of the varianceStabilizingTransformatio() function from DESeq2?

c. Other option?

ADD REPLY • link 7.1 years ago harelarik ▴ 60

score 1 · Answer 1 · 2017-12-04

1

Entering edit mode

Steve Lianoglou ★ 13k

@steve-lianoglou-2771

Last seen 22 months ago

United States

I doubt you can combine these into a single number in any meaningful way.

You need an algorithm that can take observations and their weights and use them productively together ...

ADD COMMENT • link 7.1 years ago Steve Lianoglou ★ 13k

0

Entering edit mode

Thank you!

ADD REPLY • link 7.1 years ago harelarik ▴ 60

score 0 · Answer 2 · 2017-12-05

0

Entering edit mode

Gordon Smyth 52k

@gordon-smyth

Last seen 6 hours ago

WEHI, Melbourne, Australia

I always recommend cpm(counts, log=TRUE, prior.count=3) for the purpose of other down-stream analyses, because the voom quantities cannot be summarized in single combined quantity.

This is mentioned briefly in Section 2.15 of the edgeR User's Guide. Section 15.4 of the limma User's Guide is also of interest.

ADD COMMENT • link 7.1 years ago Gordon Smyth 52k

0

Entering edit mode

Thank you!

Would you recommend using logCPM on raw counts on or TMM transformed values?

ADD REPLY • link 7.1 years ago harelarik ▴ 60

0

Entering edit mode

Yes, I usually recommend TMM. Just use cpm(dge, log=TRUE, prior.count=3), where dge is the DGEList, and the normalization factors will be used automatically.