Hi !

When  I use edgeR to find differential expressed genes between multiple conditions, I find completely different results with  glmQLFTest (I  run the test and then I filter on logFoldchange and p-value ) and glmTreat (I choose the log fold change in the argument lfc of the function )  functions .

Why such differences and which method is the good one ?

Ty for answers,

Have a good day

Have a read of this case study:

   https://f1000research.com/articles/5-1438/v2

which uses both glmQLFTest and glmTreat.

We strongly discourage the first approach that you mention, of testing and then doing posthoc filtering by fold change. That will tend to artificially favour genes at lower expression level and does not control the FDR rate. The glmTreat approach is far superior.

You might like to look at this older article that explains why the treat approach is not the same as just filtering by fold change:

  https://academic.oup.com/bioinformatics/article/25/6/765/251641

The help pages for the various functions also contain a lot of relevant information.