[BioC] P-values for tissue specific genes

Kasper Daniel Hansen khansen at stat.Berkeley.EDU
Tue Jul 15 19:45:44 CEST 2008


On Jul 15, 2008, at 12:42 AM, Pascal Gellert wrote:

> Hello,
>
> unfortunately I have I big problem I can't solve.
>
> I have to analyze if a gene is tissue specific. For example for the  
> gene xyz I have following expression values:
>
> Heart           Liver              Brain        ....several others
> 8.998497   10.013561   12.277407
> 9.743556   10.137574   11.033957
>
> For every tissue I have two values from two different experiments.
>
> Now I want to test if Heart is significant higher than Liver and  
> Brain, or Liver is significant higher than Heart an Brain, ...
>
> I read about the pairwise.t.test, but I dont't want to test Heart  
> against Liver and Heart against Brain, ...
>
> Is there any possibility to do that? ANOVA? Duncan? Constrasts?
>
> When I set up a contrasts matrix like
>
>          Heart        Liver           Brain
> C1        1             -1/2           -1/2
> C2       -1/2            1             -1/2
> C3       -1/2         -1/2              1
>
> to get tissue specific genes, some p-values are exact 0 (even I  
> corrected with holm). This can't be correct? Are only degree of  
> freedoms times contrasts allowed?

Whether or not this contrast matrix makes sense depends on your design  
matrix.

Given your p-values I would guess you are doing something wrong, but I  
have no idea what. There could also be some problems with your data  
but again it is hard to tell you what is wrong. Take a look at the raw  
data for some of the genes with a p-value of zero, perhaps you can  
learn something.

Kasper



More information about the Bioconductor mailing list