Combing Effects (t-stats) from experiment with common reference design?

Dear AK,

I assume that you have multiple replicates of each heart stage, so that 
the total number of samples is greater than 20.

To rank genes by average up-regulation in heart across all stages, simply 
form a contrast that is the average of all Stage vs Control comparisons.

For example,

  Group <- relevel(Group, ref="ControlX")
  design <- model.matrix(~Group)
  fit <- lmFit(y, design)
  cont <- c(-20,rep(1,20))/20
  fit2 <- contrasts.fit(fit, contrast=cont)
  fit2 <- eBayes(fit)
  topTable(fit2)

Best wishes
Gordon

> Date: Thu, 28 Aug 2014 21:21:03 -0400
> From: Atul <[email protected]>
> To: "Ryan C. Thompson" <[email protected]>
> Cc: "[email protected]" <[email protected]>
> Subject: Re: [BioC] Combing Effects (t-stats) from experiment with
> 	common reference design?
>
> Hi Ryan,
>
> Thanks for taking out time to reply to my question. I have samples from
> two tissues - Heart (20 different developmental stages) and Control
> (rest of the body, single fused sample from multiple time points). I
> performed 'limma' analysis (GLM approach) to identify up-regulated genes
> for each of the Heart stages (n=20).
>
> Ex comparisons:
> Heart Stage-1 vs. Control-X
> Heart Stage-2 vs. Control-X
> .....
> Heart Stage-20 vs. Control-X
>
> Now I would like to rank genes on the basis of their enrichment in heart
> across all stages. So that a gene which is highly enriched in heart
> should rank high (on top) and genes which are not enriched in heart
> should rank low (at bottom). Is there any way to combine 't-stats' for
> each stage to a single metric? Or any other method rank genes that are
> enriched in Heart across all stages?
>
> Actually I do have F-statistic. But I think that F-stat is high for gene
> which shows variable enrichment i..e gene which is not enriched in 5
> stages but enriched in 15 stages will have better F-stat reather than a
> gene with enrichment in all 20 stages. Therefore 'F-stat' doesn't seem
> to be the correct indication of enrichment level across all stages. I
> might be wrong, please correct me if that the case.
>
> Best
>
> AK
>
> On 08/28/2014 06:09 PM, Ryan C. Thompson wrote:
>> Hi Atul,
>>
>> Typically if you are testing multiple contrasts simultaneously, you
>> would use an ANOVA test that would five you an F statistics (and
>> corresponding p-value). But it's not exactly clear if that's what
>> you're asking for, Can you explain in more detail exactly which
>> hypothesis you are trying to test? Ar you trying to test whether any
>> of the Stages is different from the control, or are you trying to test
>> whether genes are changing between all Stages?
>>
>> -Ryan
>>
>> On Thu 28 Aug 2014 12:52:47 PM PDT, Atul wrote:
>>> Hi All,
>>>
>>> I was wondering whether there is any approach to combine 't-stat' from
>>> different comparisons but using same control. These are my contrasts:
>>>
>>> Stage1 vs ControlX
>>> Stage2 vs ControlX
>>> Stage3 vs. ControlX
>>> .........
>>> Stage 20 vs. ControlX
>>>
>>> Here the control is same i.e. same sample for all contrasts. From
>>> 'limma' analysis I have Fold change, t-stats and p-values for each gene.
>>>
>>> Now, is it possible to combine 't-stats' from all different stages to
>>> single value? Or compute a single combined value for all the contrasts.
>>> So, that this single metric could be used to rank genes across all time
>>> points. Is there any package available to do so? I can find methods to
>>> combine p-values but not the 't-stat'.
>>>
>>> Thanks
>>>
>>> AK

______________________________________________________________________
The information in this email is confidential and intend...{{dropped:4}}

_______________________________________________
Bioconductor mailing list
[email protected]
https://stat.ethz.ch/mailman/listinfo/bioconductor
Search the archives: http://news.gmane.org/gmane.science.biology.informatics.conductor