Deseq : p-val & p-adj variation BUT foldchange CONSTANT

[chdem]

Hi everybody !


This is my first post here, even if i'm a frequent seqanswers reader, and this time, I didn't found any answer to my question !


Firstly, i'm working on DESeq_1.10.1.

I'm trying to use deSeq from RNA-seq's data, after using mirDeep2 scripts.

I've got 84 samples and, as a good user, I would like to well understand the estimateDispertion step in order to find the best parameters, and in this way, i've made different tests with the same data file, changing estimateDispersion parameters (sharingMode="gene-est-only" and/or method="per-condition" and/or fitType="local"), and for each, write a csv data file to compare results.


And there's something that sounds strange to me :
for the same miRNA (gene in deSeq view), probabilities have changed depending on parameters BUT fold change values are always the same !

For exemple, with the mir-155 (i've exactly the same behaviour with all my miRNA) :

PARAMETERS______|MIRNA_____|__baseMean___|__baseMeanA__|_baseMeanB___|__foldChange__|__log2FoldChange|pval_______|__padj_______|
LOCAL.csv________|hsa-mir-155_|39232.49235__|53020.62099__|19766.89898__|0.372815305__|-1.423467008___|0.000592469|0.036584953_|
PER_CONDITION.csv|hsa-mir-155_|39232.49235__|53020.62099__|19766.89898 __|0.372815305__|-1.423467008___|0.000152184|0.005369934_|



Considering the parameters used, fold change must be different, isn't it ? Do I make any mistake ?

In advance, thanks a lot to the community for the help you could gave !


Chdem

[chadn737]

Fold change is dependent upon baseMeanA and baseMeanB. The estimateDispersions steps are affecting how the variance is determined. The Means are unaffected, but variance around the mean changes and this affects your pval.