Question: DESeq2 use of LRT
gravatar for soheilazareie
27 days ago by
Canada, Vancouver, University of British Columbia
soheilazareie0 wrote:


I have a question regarding RNAseq data analyses by DESeq2,

I have some data with 3 conditions (A, A1, D) and 3 other factors (P, B, Sex) with 2 or 3 levels in each,

my question is that if I use LRT without any contrast, is that correct to imply whatever results give me as a final output for comparison across these 3 levels in my condition,

I used this code: 

ddsMF <- DESeq(dds, test="LRT", reduced = ~ P + B + Sex, parallel=TRUE)
res_A_D_July17_ncon <- results(ddsMF)

2- if not, how a two-way comparison of DESeq2 is able to tell me how different the 3 conditions are from each other,

I mean if I use A vs D, A1 vs D, A vs A1, what is the big picture to see all the differences across all the 3 levels of condition.


ADD COMMENTlink modified 27 days ago by Michael Love19k • written 27 days ago by soheilazareie0
gravatar for Michael Love
27 days ago by
Michael Love19k
United States
Michael Love19k wrote:

I can't tell what that LRT does because you didn't list the full design. The LRT tests whether the coefficients that are in the full design and not in the reduced design explain more of the variability in the counts than expected under the null hypothesis. If the coefficients in the full design that are missing from the reduced design are useful in explaining the variability in the counts then the gene gets a small p-value.

The LRT will tell you if there are differences across all levels of a factor, if you remove that variable from the full design to make the reduced design. Pairwise contrasts are just telling you about differences between the pair of levels.

ADD COMMENTlink written 27 days ago by Michael Love19k

Hi Michael,

Thanks for the reply, I would appreciate your help.

our question is that considering 3 factors other than condition (3 levels, A, A1, D) which includes sex, passage of cells and different batches of expermint, what are the differntially expressed genes across all our samples (35, 14 D, 8 A1, and 13 of A). how different our samples are in terms of gene expression, 

here is my full design,

dds <- DESeqDataSetFromMatrix ( countData = cts,
                                colData = sampleinfo_July17,
                                design = ~ B + P + Sex + condition )

ddsMF <- DESeq(dds, test="LRT", reduced = ~ P + B + Sex, parallel=TRUE)
(is this code necessary for looking at all the samples? as its just paired-wise comaprison) res_A_D_July17 <- results(ddsMF, contrast=c("condition","A","D"))
mcols(res_A_D_July17, use.names = TRUE)
summary(res_A_D_July17) # summarize some basic tallies
resultsNames(ddsMF) # lists the coefficients

resultsNames(ddsMF) # lists the coefficients
[1] "Intercept"         "B_B2_vs_B1"        "B_B3_vs_B1"       
[4] "B_B4_vs_B1"        "P_R_vs_q"          "Sex_M_vs_F"       
[7] "condition_A_vs_D"  "condition_A1_vs_D"

ADD REPLYlink modified 27 days ago • written 27 days ago by soheilazareie0

So getting back to the first question, yes the LRT here is testing across all levels of the condition variable.

See the sections of the help page ?results and the FAQ in the vignette about why only one coefficient is shown with the LRT.

You can see all the coefficients in the model with coef(dds) but the LRT represents a test of multiple coefficients here.

ADD REPLYlink written 27 days ago by Michael Love19k
Please log in to add an answer.


Use of this site constitutes acceptance of our User Agreement and Privacy Policy.
Powered by Biostar version 2.2.0
Traffic: 323 users visited in the last hour