DESeq2 test LRT or WALD
Entering edit mode
Last seen 4.6 years ago

Dear Mickael,

I would like to use DESeq2 to do a differential analysis (genus level) and I contact you concerning the use of the test LRT or wald.

Design of the study : Metagenomic 16S data 4 timepoints 2 groups : group with product (P) and group with control product (CP) Dose : group with 3 doses and group with 1 dose per day Thus I have : P.1 / P.3 / CP.1 / CP.3 at 4 timepoints In the tested product (P), there is a bacterial genus which is not present in the CP product. It is considered as our positive control.

I tested a global differential analysis with P.1 / P.3 / CP.1 / CP.3 at 4 timepoints and a targeted differential analysis with P.3 / CP.3 at 2 timepoints (baseline and interest timepoint)

I didn’t obtain the same results with the test LRT and the test Wald. more genus significant with the LRT test and especially our positive control genus which is significant (p value adjusted) with the LRT test (padjust=1.15e-05 ) and not significant with the wald test (padjust=0.83) The logFoldChange of the positive control genus is higher with the dose 3 than the dose 1 in the case of the LRT test ( what seems to be consistent) But the LFC is lower with the dose 3 than the dose 1 in the case of the wald test.

Is it possible to use the test LRT in this case ? Is it statistically correct ? Or not and it is preferable to use wald test ?

I saw in the forum : “the wald (applicable when we have two condition) and LRT test (when we have more than 2 conditions and intent to check interaction between conditions)” Is it right ? could you explain me

Thanks for your quicky answer and help.

Best regard Anne-Sophie

deseq2 LRT WALD • 1.8k views
Entering edit mode
Last seen 3 hours ago
United States

I'm not convinced that DESeq2 is the best tool for metagenomic data. I know groups have shown that it may provide reasonable output on some datasets, but I've never analyzed metagenomic data myself, and the tool hasn't been designed or adapted to work with data of the distribution seen in metagenomics or microbiome. That said, I've found that the LRT performs better than Wald on data that is sparser than the NB can accommodate. So if you were to use DESeq2, I'd recommend the LRT over the Wald.


Login before adding your answer.

Traffic: 648 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