Question

LRT test multiple comparisons

0

Entering edit mode

laianavarromartin ▴ 20

@laianavarromartin-7750

Last seen 7.0 years ago

Spain

Hi,

My question refers to the multiple comparison done by LRT test. I though that this test would compare all the levels of a factor between them, but if I'm not mismatching...it is only comparing all the levels with the one I'm considering the "control" group. This is my experimental design:

Condition	Replicate	SampleName
Control	R1	AB3131
0.03ppm	R1	AB3132
0.3ppm	R1	AB3133
1ppm	R1	AB3134
Control	R2	AB3139
0.03ppm	R2	AB3140
0.3ppm	R2	AB3141
1ppm	R2	AB3142
Control	R3	AB3143
0.03ppm	R3	AB3144
0.3ppm	R3	AB3145
1ppm	R3	AB3146

I run dds as follows:

dds <- DESeqDataSet(se, design = ~ Replicate + Condition)
dds$Condition <- relevel(dds$Condition, "Control")
ddsLRT <- DESeq(dds, test="LRT", full= ~ Replicate + Condition,  reduced = ~ Replicate)

and this are the results:

resLRT <- results (ddsLRT)

log2 fold change (MLE): Condition 1ppm vs Control
LRT p-value: '~ Replicate + Condition' vs '~ Replicate'
DataFrame with 26873 rows and 6 columns

Now i'm interested to extract all the Log2FC of all comparisons, so I checked the results names to see the comparisons done by dds

resultsNames(ddsLRT)

I get the following:

[1] "Intercept" "Replicate_R2_vs_R1" "Replicate_R3_vs_R1"
[4] "Condition_0.03ppm_vs_Control" "Condition_0.3ppm_vs_Control" "Condition_1ppm_vs_Control"

so does that mean that the test did not compared for example condition 1ppm? vs 0.3ppm?

deseq2 extracting contrasts • 1.8k views

ADD COMMENT • link 7.8 years ago laianavarromartin ▴ 20

0

Entering edit mode

Hi Mike,

First of all thanks for all your patience and time dedicated to us. I think I have been mistaken these tests all the time. I have an appointment with a local statistician hoping to help me on this...but I had in my mind that doing the Walt test was similar to a t-test, and doing a LRT was similar to do an ANOVA. So is this statement wrong? I guess all the confusion comes from here...

ADD REPLY • link 7.8 years ago laianavarromartin ▴ 20

score 0 · Answer 1 · 2017-02-21

hi,

I think you would benefit from meeting with a local statistician to better understand the motivation of the likelihood ratio test and how to interpret the test result. This is discussed a bit in ?results under the paragraph about likelihood ratio test, and also in the FAQ in the vignette, under the question "I ran a likelihood ratio test, but results() only gives me one comparison..."