Question

Problem in designing result in DESeq2

0

Entering edit mode

archana • 0

@archana-15349

Last seen 6.0 years ago

United Kingdom

I am trying to analyse RNASeq data with following design:

dds <- DESeqDataSetFromMatrix(countData = dataCountTable, colData = dataDesign, design = ~ genotype+treatment+genotype:treatment)

dds <- DESeq(dds)

resultsNames(dds)

[1] "Intercept" "genotype_F_vs_C" "genotype_M_vs_C"

[4] "treatment_SAP_vs_GFP" "genotypeF.treatmentSAP""genotypeM.treatmentSAP"

dataDesign

genotype treatment

sample1 F GFP

sample2 F GFP

sample3 F GFP

sample4 F GFP

sample5 F SAP

sample6 F SAP

sample7 F SAP

sample8 F SAP

sample9 M GFP

sample10 M GFP

sample11 M GFP

sample12 M GFP

sample13 M SAP

sample14 M SAP

sample15 M SAP

sample16 M SAP

sample17 C GFP

sample18 C GFP

sample19 C GFP

sample20 C GFP

sample21 C SAP

sample22 C SAP

sample23 C SAP

The question I am trying to answer is:

1. Which genes are DE in male-exposed SAP54 versus male-exposed GFP plants, that are not DE in other comparisons. Meaning these genes should be not DE or significantly less DE or DE in the opposite direction in female-exposed GFP versus female-exposed SAP54 plants, or in male-exposed GFP or non-exposed GFP plants.

2.What does these interaction means genotypeF.treatmentSAP and genotypeM.treatmentSAP?

Any help would be appreciated Thanks!

deseq2 • 638 views

ADD COMMENT • link updated 6.1 years ago by Michael Love 42k • written 6.1 years ago by archana • 0

Michael Love · Answer 1 · 2018-03-29

0

Entering edit mode

Michael Love 42k

@mikelove

Last seen 14 hours ago

United States

Please take a look at the vignette, which has a diagram of how to interpret interaction terms.

Here the main effect of treatment is SAP vs GFP for the samples where genotype is C. The interaction terms represent the difference -- how much higher or lower the treatment effect is -- when genotype is F or M.

You don't have a single term or combination of terms that represents something like "treatment effect is significant for genotype M but not the other genotypes".

What you could do is instead fit a model: ~genotype + genotype:treatment, which will give you three separate treatment effects. You can build three results tables for the treatment effect for C, F, and M, and then look at those genes which have adjusted p < some threshold for M but not for C and F.

ADD COMMENT • link 6.1 years ago Michael Love 42k

0

Entering edit mode

Thanks Michael. When I do the suggested condition, I get following:

dds <- DESeqDataSetFromMatrix(countData = dataCountTable, colData = dataDesign, design = ~genotype + genotype:treatment)

resultsNames(dds)

[1] "Intercept" "genotype_F_vs_C" "genotype_M_vs_C"

[4] "genotypeC.treatmentSAP" "genotypeF.treatmentSAP" "genotypeM.treatmentSAP"

If I understand correctly "genotypeC.treatmentSAP" : Shows SAP effect on genotypeC

"genotypeF.treatmentSAP" : shows SAP effect on genotypeF

"genotypeM.treatmentSAP" : shows SAP effect on genotypeM

And if I look for genes which are present in genotypeM.treatmentSAP and not in other I can get SAP effect specific to Male. But These genes might be present in genotypeM.treatmentGFP as well. But I am interested in genes which are specific to GFP treatment and should not be there in GFP?

Looking forward for your comment . Thanks!

ADD REPLY • link updated 6.1 years ago by Michael Love 42k • written 6.1 years ago by archana • 0

0

Entering edit mode

There is no meaning to "genotypeM.treatmentGFP" in this design.

What you have, genotypeM.treatmentSAP, gives you the treatment effect (SAP vs GFP) for genotype M.

I'm suggesting to build three results tables, using results(dds, name="genotypeM.treatmentSAP") etc.

Then find the set of genes that are in the M comparison but not in the others. This is the best you can do, there isn't a single contrast that will produce the set that you're looking for.