[BioC] limma_analysis result

Wed Jun 18 07:51:55 CEST 2008

Hi Abhilash,
In addition to Francois' comments, for me the biggest indicator is  
that your adjusted P-values are all greater than 0.05.
My interpretation of this is that after multiple testing correction,  
none of your genes are statistically significantly differentially  
expressed.
This probably implies that either the differences between your two  
groups are not large, or that there is higher inter-sample variance;  
again, plotting some of these DE genes will help inform you as to  
which is the case.

cheers,
Mark

On 18/06/2008, at 2:48 AM, Francois Pepin wrote:

> Hi Abilash,
>
> you might want to read up a bit on those statistics and how they are  
> generated. It definitely helps when it comes to interpreting their  
> results properly.
>
> The B statistic is a log odds ratio. It simply means that the actual  
> log odds are between 0 and 1: less than 50% probability of  
> differential expression according to this test.
>
> The t statistics can be positive or negative. A negative t-statistic  
> simply means that the mean of the second group ("test" in this case)  
> is higher. A high negative t-statistic would have the same evidence  
> of differential expression as a high positive t-statistic.
>
> Francois
>
> Abhilash Venu wrote:
>> Hi Mark,
>> I think it was a great suggestion and I am providing the result,  
>> which I got
>> after the command, topTable(fit, coef="normvstest", adjust="fdr").
>> I believe this preliminary results solved the problem of very high  
>> fold
>> value, which I was getting earlier. I will be looking at the entire  
>> data and
>> proved a better view at the earliest. But I have some doubts in  
>> this table
>> itself. Here I am getting negative odds ratio and in some cases  
>> negative t
>> value. What should I do in these scenario?
>> topTable(fit, coef="normvstest", adjust="fdr")
>>           logFC   AveExpr         t      P.Value adj.P.Val         B
>> 35726 -2.1103554 11.936825 -9.602581 8.028641e-06 0.3614093 -3.912633
>> 1968  -1.3413791  9.974470 -6.960746 9.138620e-05 0.7305539 -3.971277
>> 5558  -1.6566417 10.885625 -6.506724 1.487751e-04 0.7305539 -3.987395
>> 34497  1.0445013 10.251047  6.185219 2.132063e-04 0.7305539 -4.000460
>> 33195 -1.3603874 13.373106 -6.116817 2.305481e-04 0.7305539 -4.003438
>> 44662  0.9528248 11.180259  6.045345 2.503347e-04 0.7305539 -4.006630
>> 24980 -1.5689151 10.824414 -5.932376 2.855094e-04 0.7305539 -4.011846
>> 30206  2.2991372 13.647875  5.926758 2.873946e-04 0.7305539 -4.012112
>> 26046 -1.1709614  9.505652 -5.746545 3.557246e-04 0.7305539 -4.020911
>> 27210  1.4815342  9.416698  5.656415 3.964086e-04 0.7305539 -4.025537
>> Thanks in advance
>> Best
>> Abhilash
>> On Tue, Jun 17, 2008 at 4:38 AM, Mark Cowley <m.cowley0 at gmail.com>  
>> wrote:
>>> Hi Abhilash,
>>> Your code looks good, except that usually you will want to  
>>> normalise log
>>> transformed data. thus try:
>>>
>>>> MA<-normalizeBetweenArrays( log2(Rgene$G), method="quantile")
>>>>
>>> If your logFC ratios still look very high, then try convincing  
>>> yourself of
>>> their accuracy by looking at the raw data (RG$R) for some of the  
>>> most
>>> differentially expressed genes, and also plot the expression  
>>> values for some
>>> of these DE genes.
>>>
>>> good luck,
>>> Mark
>>> Peter Wills Bioinformatics Centre
>>> Garvan Institute of Medical Research
>>>
>>>
>>>
>>> On 17/06/2008, at 1:17 AM, Abhilash Venu wrote:
>>>
>>> Hi list,
>>>> I am still wonder about the data, which I analyzed by the limma.  
>>>> I accept
>>>> that  I am a biology graduate student, and in the learning stage.  
>>>> I am
>>>> analyzing the single color data, which had been generated by  
>>>> Agilent 4x44k
>>>> platform. With the help of mailing list and limma users guide, I  
>>>> have done
>>>> the following analysis. But logFC gives very high values like  
>>>> 320, 1320
>>>> etc.
>>>> I don't know how really the fitting is happening. Can I rely on  
>>>> this
>>>> result.
>>>> How should I go about it.
>>>> #Reading the data.
>>>>
>>>> RG<-read.maimages(txt_files, columns = list(G = "gMeanSignal", Gb =
>>>>> "gBGMeanSignal",
>>>> R="gMedianSignal",Rb="gBGMedian
>>>>
>>>>> Signal"),
>>>>> annotation= c("Row", "Col",
>>>>> "ProbeUID","ProbeName", "GeneName",))
>>>>>
>>>>
>>>> Rgene<-backgroundCorrect(RG,method='subtract')
>>>>
>>>> #Considering only G as it is single color experiment.
>>>> MA<-normalizeBetweenArrays(Rgene$G,method="quantile")
>>>>
>>>> design <- cbind(norm=1,normvstest=c(1,1,1,1,0,0,0,0))
>>>> fit <- lmFit(MA, design)
>>>> fit <- eBayes(fit)
>>>> topTable(fit, coef="normvstest", adjust="fdr")
>>>> --
>>>>
>>>> Regards,
>>>> Abhilash
>>>>
>>>>       [[alternative HTML version deleted]]
>>>>
>>>> _______________________________________________
>>>> 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
>>>>
>>>
>>>
>>> ----------------------------------------------------------------------
>>> Mark Cowley, BSc (Bioinformatics)(Hons)
>>>
>>> Peter Wills Bioinformatics Centre
>>> Garvan Institute of Medical Research
>>> 384 Victoria St                                 Tel:  +61 2 9295  
>>> 8542
>>> Darlinghurst, NSW 2010          Fax:  +61 2 9295 8538
>>> Australia                                       email:
>>> m.cowley at garvan.org.au
>>> www.garvan.org.au
>>> ----------------------------------------------------------------------
>>>
>>>
>