[BioC] valcano plot x-values

[James W. MacDonald]

sabrina.shao wrote:
> 
>>> Hi, Jim:
>>> Thank you very much for the explaination. So basically if I do the 
>>> Ftest, the volcano plot won't tell me much in terms of fold changes 
>>> of each treatment vs. control ( assume that I have one control and 3 
>>> treatments). In my case, what is the best way to illustrate the 
>>> average fold changes withought including he differences between any 
>>> pairs? Thanks!
>>
>>
>>
>> I don't understand the question. Are you asking how to produce a 
>> volcano plot without specifying which comparisons to plot?
>>
>> Note that you *can* produce volcano plots for each comparison by 
>> specifying test.type = "ttest" in your call to matest(). You will have 
>> to specify a contrasts matrix in this case so matest() knows what 
>> comparisons you are interested in.
>>
> Hi, Jim:
> No, I am not saying to make a volcano plot without specifying which 
> comparisons to plot. I am asking if and how I can make a volcano plot 
> with " averaged" Fs results from comparisons with all treatments vs.the 
> only control. It is similar to multiple t-test , I guess, but not 
> exactly the same since all treatments are compared to the same control. 
> My main interest is to find out if there is any significant difference 
> between the treatment vs. control, not the difference among treatments. 
> :(  I wonder how t-test can work out since I only have 3 samples per 
> treatment, with such a small sample size, paired t-test would be 
> unstable /unreliable. Any suggestions?

Are these samples paired? If so, you should not be ignoring this fact. 
Although you will be 'losing' degrees of freedom you gain much more by 
accounting for the dependence between the paired samples. I don't know 
if maanova can handle paired data - you might be better off using limma.

Anyway, to do a volcano plot with all treatments vs control, you will 
need to do matest() and specify a contrasts matrix as well as 
test.type="ftest". The contrasts matrix will have to be specified to 
make all the treatment vs control comparisons. Unfortunately I don't 
have the time to get into how one would do such a thing. The example in 
?matest shows how to specify one such matrix, so that may be of some 
help to you.

> 
> Also I got your reply on the q-value issue. I know I am confused. I 
> looked that example from adjPval and could not figure out how it was 
> adjusted and what method it uses . But for q-value, as far as I 
> understand, it is the lower bound of FDR. So should I convert the 
> adjusted P values to q-values or just convert the original p-values from 
> any of the Ftest statistics to the q-value?

I think the main difference between the q-value (I assume from the 
qvalue package) is that the q-value is based on pFDR where you are 
assuming at least some genes are actually differentially expressed, 
whereas the p-values from adjPval() are based on different versions of 
Benjamini and Hochberg (or Yekutieli) FDR where there is (AFAIK) no such 
assumption. Although there are some differences between these methods, I 
don't think the resulting gene lists will be so different that you need 
to worry about which method you use.

Best,

Jim


> 
> Thanks!
> 
> Sabrina


-- 
James W. MacDonald, M.S.
Biostatistician
Affymetrix and cDNA Microarray Core
University of Michigan Cancer Center
1500 E. Medical Center Drive
7410 CCGC
Ann Arbor MI 48109
734-647-5623


**********************************************************
Electronic Mail is not secure, may not be read every day, and should not be used for urgent or sensitive issues.