Question: In DESeq2, why LRT result's LFC doesn't match with any of Wald test result's LFC?
gravatar for gmhhope
2.2 years ago by
gmhhope0 wrote:

Hi DESeq2,

I have a question with respect to DESeq2.

I have one factor with 4 levels (wild-type, mutant A, mutant B, mutantA&B; the wild-type is set as reference). Based on this, I want to do LRT test to get analysis just like ANOVA and then compare two by two among them Because there are no post-hoc test, so I also did multiple pairwise comparison using Wald test. I learned from the tutorial that the LRT test will give only one of the Log2FC among all others. So to confirm that, I try to match the Log2FC given by LRT test's result table with those other Log2FC given by Wald test. However, I just cannot find any of them fully matched. I found one which is very similar. I guessed it may be that; however, I want to know why will this happen? And I found difficult to understand how the output of coef() translate into LFC? Can anybody help explain to me? 


The R code is shown below:

dds <- DESeqDataSetFromMatrix(countData = countData, colData = colData, design = ~class)

dds <- dds[rowSums(counts(dds))>1,]

dds$class <- relevel(dds$class, ref="P.UF1")
### Perform DESeq function to get all MLE & bayes procedure (including MAP)
dds1 <- DESeq(dds,test="LRT",reduced=~1)
dds2 <- DESeq(dds)
### Perform results function to get the statistics and LFC
res_LRT <- results(dds1) 
res_DmmcAvsPUF1 <- results(dds2,contrast=c("class","DmmcA","P.UF1"))
res_DpduPvsPUF1 <- results(dds2,contrast=c("class","DpduP","P.UF1"))
res_DmmcA_pduPvsPUF1 <- results(dds2, contrast=c("class","DpduP_mmcA","P.UF1"))
res_DmmcAvsDpduP <- results(dds2, contrast=c("class","DmmcA","DpduP"))
res_DpduP_mmcAvsDmmcA <- results(dds2, contrast=c("class","DpduP_mmcA","DmmcA"))
res_DpduP_mmcAvsDpduP <- results(dds2, contrast=c("class","DpduP_mmcA","DpduP"))


The 2 columns representing similar log2FCs is shown below.

LRT_log2FoldChange MPvsPUF1_log2FoldChange
0.631377697 0.627936831
1.051816895 0.898240504
-0.253159341 -0.246927052
-0.010568974 -0.010054782
-0.788608064 -0.78341221
-0.092539348 -0.090469655
-0.692334811 -0.687365411
0.503294525 0.456056859
-0.441611727 -0.438624489


ADD COMMENTlink modified 2.1 years ago by Michael Love20k • written 2.2 years ago by gmhhope0
gravatar for Michael Love
2.1 years ago by
Michael Love20k
United States
Michael Love20k wrote:

You are showing the moderated LFC using the Wald test and the unmoderated LFC using the LRT. The LRT uses an unmoderated fold change, while it is your choice whether to moderate or not (betaPrior=TRUE or FALSE) for the Wald test.

This information (whether you produced MAP or MLE log fold change) is printed at the top of the results table if you print it:

> res
ADD COMMENTlink written 2.1 years ago by Michael Love20k
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: 242 users visited in the last hour