[BioC] vennDiagram Statistics Advice

[Noah Dowell]

Thank you Wolfgang!!

The Fisher test looks like it will work nicely.  I will try to get my hands on the Agresti book to make sure I am using it correctly.

I will take another stab at a heatmap approach on specific subsets.  My initial attempts at clustering and heatmap presentation were less than convincing.  My statistician friends suggested it was due to less than robust data and my biologist friends suggested that it was inherent to looking at subtle (single point mutants versus complete knock outs) biological effects.  Hopefully we can all meet somewhere in the middle:)

Best,

noah


On Apr 27, 2010, at 4:56 AM, Wolfgang Huber wrote:

> 
> Dear Noah
> 
> a first, immediate answer is to look at
> ? fisher.test
> 
> Some light reading on its background can be found here: http://en.wikipedia.org/wiki/Fisher%27s_exact_test
> and see e.g. Agresti's excellent book 'Categorial Data Analysis' for something more substantial.
> 
> The second, more conceptual and perhaps more useful answer is to use heatmaps and clustering of the expression profiles to show the desired similarity relationships between the 4 expression profiles.
> 
> Simply using Euclidean distance, say on the subset of genes that is the union of all differentially expressed genes, on the fold changes should be a good start.
> 
> 	Best wishes
> 	Wolfgang
> 
> 
> Noah Dowell scripsit 27/04/10 00:37:
>> Dear All,
>> I have used the excellent limma package to analyze my 2-color Agilent Yeast microarray data and have determined the differentially expressed genes (as compared to wild-type expression) in four independent experiments.  I am showing the summary of my results below:
>>> results <- decideTests(fit2, method="global")
>>> summary(results)
>> # mutant1 mutant2 mutant3  deletionstrain
>> #-1       147       126        40        252
>> # 0       5924      6033      6171       5600
>> # 1        185        97        45        404
>> The three mutant experiments represent expression data from cells expressing different point mutants in the gene that is deleted in the expression strain.  The point mutation in Mutant3 is a control that should not affect the protein's function given our current knowledge of how the protein works therefore the relatively small number of genes differentially expressed as compared to wild-type is consistent with our using that strain as a "control."
>> Mutants1 and 2 are similar in their biological defects.  They are also hypomorphic alleles as compared to the complete deletion strain so the smaller number of differentially expressed genes is again consistent with our working model.
>> I have created vennDiagrams of these four experiments to look at the overlap of differential expression between experiments.
>> My question is if there is a test I can run on the vennDiagrams (or simply on the overlapping gene lists) to show that there is significantly more overlap between mutant 1 or mutant 2 and the deletion strain when compared to mutant 3?
>> Thank you for your time and input!
>> Noah
>> _______________________________________________
>> Bioconductor mailing list
>> Bioconductor at stat.math.ethz.ch
>> https://stat.ethz.ch/mailman/listinfo/bioconductor
>> Search the archives: http://news.gmane.org/gmane.science.biology.informatics.conductor
> 
> 
> -- 
> 
> 
> Wolfgang Huber
> EMBL
> http://www.embl.de/research/units/genome_biology/huber
> 
>