[BioC] Venn Diagrams in Limma

[Gordon Smyth]

At 10:23 PM 28/10/2003, Jason Skelton wrote:
>Gordon Smyth wrote:
>>Venn diagrams are by their nature simply overall counts. But you can 
>>easily identify the relevant genes from the 'classification' object from 
>>which the Venn diagram is computed.
>
>Not sure how this is done, the object I have created in ClassifyTests has 
>classification (0 and 1's) and Fstats associated with it
>am I looking in the wrong place ?  I have a list of genes made from 
>uniquegenelist is this the list I'm wanting to access ?
>My VennCounts object is:
>
>     DIF1 DIF2 DIF3 Counts
>[1,]    0    0    0   1925
>[2,]    0    0    1     18
>[3,]    0    1    0     26
>[4,]    0    1    1      5
>[5,]    1    0    0     70
>[6,]    1    0    1     13
>[7,]    1    1    0     90
>[8,]    1    1    1     59
>attr(,"class")
>[1] "VennCounts"
>
>could you give an example of how to associate each of the 8 counts with 
>genes from the uniquegenelist (OR other)

Go one step back, to the classification matrix which comes out of 
classifyTests(). That gives you a classification matrix with one row for 
each gene in the uniquegenelist.

>>2)
>>I'd also like to be able to use the false discovery rate option that can 
>>beused in classifyTestsP method for my venn diagram but I also want to 
>>use data that I've received from makeContrasts and contrast.fit which I 
>>can only use in classifyTests but I can't use "fdr" in that function....
>>
>>Is there a particular reason why I can't do this ?
>
>Yes there is. classifyTests() uses a method intended to control false 
>discovery rate across contrasts. This is not compatible with the simple 
>p.adjust() approach to controling FDR across genes. I suggest that you 
>simply find what unadjusted p-value corresponds to your desired FDR level 
>and enter that to classifyTests().
>
>OK that makes more sense now thanks.
>
>I have yet more questions about the heatdiagram option
>My gene names are very long 28+ characters long is there a heatdiagram 
>specific command that alters the size of the box that the heatmap covers ?
>So that genes appearing on the plot are not truncated (I've tried some of 
>the R commands for Margins etc but they don't want to work with high level 
>plots)

No there isn't. imageplot truncates names to 15 characters. You would have 
to modify the code yourself to change that.

>Also in the heatdiagram you can specify EB$t and fit$coef
>e.g. heatdiagram(newDIF123SEPcontrastfitEB$t, 
>newDIF123SEPcontrastsfit$coef etc)
>you can specify the critical.primary argument  which  is the critical 
>value "above" which genes are considered significant.
>
>BUT
>
>If you want to use EB$p.value fit$coef
>e.g. heatdiagram(newDIF123SEPcontrastfitEB$p.value, 
>newDIF123SEPcontrastsfit$coef etc)
>you can't specify the critical.primary because you need to specify a 
>maximum value e.g(0.05) rather than minimum as your cutoff

Use heatDiagram() instead. That function will accept a classification 
matrix from classifyTestsP which will allow you to specify your desired 
p-value cutoff.

Gordon

>Thanks for you help
>
>Jason