DESeq2 model with covariates
Entering edit mode
Last seen 3.8 years ago

Dear Michael,

i have a question regarding the usage of covariates in DESeq2 modelling. I am thinking about including a covariate, e.g. gender.

1. The model with covariate gender:

condition <- factor(c("CTRL","CTRL","CTRL","CTRL","Disease","Disease","Disease","Disease"))

sex <-factor(c("Male","Male","Female","Female","Male","Male","Female","Female"))

dds <- DESeqDataSetFromMatrix(countData = mycountdata, colData= mymatrix, design =(~ sexMale + conditionDisease))

2. The simpler model without covariate:

colData <- data.frame(condition= factor(c("CTRL","CTRL","CTRL","CTRL","Disease","Disease",

dds2 <- DESeqDataSetFromMatrix(countData = mycountdata, colData, formula(~ condition))

My question: Is there a function in DESeq2 to compare if my model including gender explains (significantly) more variance than my simple model, to make sure that i actually should include this covariate?

Best regards,Markus


deseq2 covariates • 3.2k views
Entering edit mode
Last seen 23 hours ago
United States

I don't usually use statistical testing to decide whether to include a covariate. I typically just include it, where sex as an influence on gene expression for at least some genes seems like a safe assumption. You can also examine a PCA plot using ntop=500 or higher number of genes like 1000 or 2000, to see if the samples cluster by that covariate.

If you did want to test to find genes related to sex, you could build a LRT: full=~condition + sex compared to reduced=~condition. The results table should help you find genes where sex influences gene expression.

Entering edit mode

I have a follow-up question on this. I have done this and gotten many genes where sex influences expression, but what I am interested in is how sex influences expression in a given condition. So I don't care about the Normal condition, ideally in the normal condition expression would not be significantly different. I am interested in genes whose expression is different in the test condition in one gender but not the other. Basically gender-specific differences in gene expression only in the test condition. Is there an easy way to do with with LRT?

Thanks, Joe

Entering edit mode

You can just do a test of the condition in normal using altHypothesis=“lessAbs”. You can obtain this by asking results() for the main effect (which is the condition effect for the reference level of other variable). You’d then just combine the genes that are differential across condition and have evidence of not changing in normal by using the intersect() function.


Login before adding your answer.

Traffic: 318 users visited in the last hour
Help About
Access RSS

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

Powered by the version 2.3.6