[BioC] minfi F-test Results

Dario Strbenac D.Strbenac at garvan.org.au
Thu Nov 3 07:00:09 CET 2011


Hello,

I notice from the vignette CpG plot and from looking at the raw numbers that the top two results are sites which have all scores 0 in one state, and slightly above 0 in another.

# States : A, A, B, B, A, B.
> getBeta(mset)["cg10805483", ]
5723646052_R02C02 5723646052_R04C01 5723646052_R05C02 5723646053_R04C02 5723646053_R05C02 5723646053_R06C02 
       0.00000000        0.00000000        0.04224709        0.03153657        0.00000000        0.03587444 
> getBeta(mset)["cg20386875", ]
5723646052_R02C02 5723646052_R04C01 5723646052_R05C02 5723646053_R04C02 5723646053_R05C02 5723646053_R06C02 
       0.01969133        0.02250352        0.00000000        0.00000000        0.02604167        0.00000000

I suppose this is because there is zero variance in one state and not much in the other. But, intuitively, the CpG sites shouldn't be at the top of the differentially methylated list.

Same for probes near 1 :

           intercept         f         pval       qval
cg07485273  4.273456  3630.021 4.545020e-07 0.02084478

> getBeta(mset)["cg07485273", ]
5723646052_R02C02 5723646052_R04C01 5723646052_R05C02 5723646053_R04C02 5723646053_R05C02 5723646053_R06C02 
        0.9535110         0.9443498         1.0000000         1.0000000         0.9540892         1.0000000

Are there more suitable tests to use for this data of proportions ?

--------------------------------------
Dario Strbenac
Research Assistant
Cancer Epigenetics
Garvan Institute of Medical Research
Darlinghurst NSW 2010
Australia



More information about the Bioconductor mailing list