[BioC] topTable threshold on p-value and logFC [Re: was design matrix]

Marcus Davy mdavy at hortresearch.co.nz
Mon Oct 8 03:51:44 CEST 2007


Thanks for the information.

Yes you are correct, the code;

>>   if( any(is.na(toSub)) ){
>>     toSub <- toSub[!is.na(toSub)]
>>   }
>>   return(tt[toSub, ])

is a bug and needs to be removed because of recycling it will stuff the
returned index up. That was a quick hack I added recently without thinking
about enough which is incorrect as your have pointed out.


Marcus



On 8/10/07 2:31 PM, "Martin Morgan" <mtmorgan at fhcrc.org> wrote:

> Hi Marcus -- A few comments below, for what it's worth...
> 
> Marcus Davy <mdavy at hortresearch.co.nz> writes:
> 
>> Additionally to decideTests(), I made a function which is useful for making
>> *any* filter you like. The example provided filters the same as
>> decideTests().
>> You must correctly specify the columns of interest in the ''filter''
>> expression argument so some knowledge of limma's data structures is
>> required.
>> 
>>  ttFilter <- 
>> function (filter = "abs(logFC) > 0.69 & abs(t) > 2", fit, sort.by = "t",
>>             number = nrow(fit), ...)
>> {
>>   tt <- topTable(fit, sort.by = sort.by, number = number, ...)
> 
> from here...
> 
>>   tCols <- colnames(tt)
>>   e <- new.env()
>>   for (i in tCols) {
>>     e[[i]] <- tt[[i]]
>>   }
>>   toSub <- eval(parse(text = filter), envir = e)
> 
> ... to here copies the data frame returned by topTable into an
> environment, to be used in eval. However, the 'envir' argument to eval
> can be a data.frame (!, see the help page for eval), so you could have
> just
> 
>   toSub <- eval(parse(text=filter), tt)
> 
> 'with' provides a kind of user-friendly access to this for interactive use
> 
>   toSub <- with(tt, abs(logFC) > 0.69 & abs(t) > 2)
> 
>>   if( any(is.na(toSub)) ){
>>     toSub <- toSub[!is.na(toSub)]
>>   }
>>   return(tt[toSub, ])
> 
> reducing the length of toSub (by deleting the NA's) will likely lead
> to unexpected recycling of the subscript index, e.g.,
> 
>> df <- data.frame(x=1:3)
>> df[c(TRUE,FALSE),, drop=FALSE]
>   x
> 1 1
> 3 3
> 
> Martin
> 
>> }
>> 
>> Some Examples;
>>      library(limma)
>>      set.seed(1)
>>      MA <- matrix(rnorm(100, 0,3), nc=4)
>>      fit <- lmFit(MA)
>>      fit <- eBayes(fit)
>>      topTable(fit)
>>      # Post filter on |M|>2
>>      ttFilter(filter = "abs(logFC)>2", fit)
>>      # |M|>1.4 & abs(t) > 1.8
>>      ttFilter(filter = "abs(logFC)>1.4 & abs(t)>1.8", fit)
>> 
>> 
>> Marcus
>> 
>> On 5/10/07 1:58 PM, "Gordon Smyth" <smyth at wehi.edu.au> wrote:
>> 
>>> I have changed the subject line to something more appropriate.
>>> 
>>> In R 2.5.1 and Bioconductor 2.0, the recommended way to do what you
>>> want (select DE genes on the basis of a combination of p-value and
>>> log fold change) was to use decideTests(). In R 2.6.0 and
>>> Bioconductor 2.1, you will find that topTable() now has p-value and
>>> logFC arguments which allow you to do the same thing using topTable().
>>> 
>>> Best wishes
>>> Gordon
>>> 
>>>> Date: Wed, 3 Oct 2007 17:31:34 +0100 (BST)
>>>> From: Lev Soinov <lev_embl1 at yahoo.co.uk>
>>>> Subject: Re: [BioC] design matrix
>>>> To: "James W. MacDonald" <jmacdon at med.umich.edu>
>>>> Cc: bioconductor at stat.math.ethz.ch
>>>> Message-ID: <412385.24484.qm at web27908.mail.ukl.yahoo.com>
>>>> Content-Type: text/plain
>>>> 
>>>> Dear List,
>>>> 
>>>>   Could you help me with another small issue?
>>>>   I usually write out the results of my analysis using the
>>>> write.table function as follows:
>>>> 
>>>>   Assuming 30000 probes in the dataset:
>>>>   data <- ReadAffy()
>>>>   eset <- rma(data)
>>>> 
>>>>   design <- model.matrix(~ -1+factor(c(1,1,1,2,2,3,3,3)))
>>>>   colnames(design) <- c("group1", "group2", "group3")
>>>>   contrast.matrix <- makeContrasts(group2-group1, group3-group2,
>>>> group3-group1, levels=design)
>>>> 
>>>>   fit <- lmFit(temp, design)
>>>>   fit2 <- contrasts.fit(fit, contrast.matrix)
>>>>   fit2 <- eBayes(fit2)
>>>> 
>>>>   C1<-topTable(fit2, coef=1, number=30000, adjust="BH")
>>>> 
>>>> write.table(C1,file="comparison1.txt",append=TRUE,quote=FALSE,sep="\t",row.
>>>> na
>>>> mes=TRUE,col.names=FALSE)
>>>> 
>>>>   C2<-topTable(fit2, coef=2, number=30000, adjust="BH")
>>>> 
>>>> write.table(C2,file="comparison2.txt",append=TRUE,quote=FALSE,sep="\t",row.
>>>> na
>>>> mes=TRUE,col.names=FALSE)
>>>> 
>>>>   C3<-topTable(fit2, coef=3, number=30000, adjust="BH")
>>>> 
>>>> write.table(C3,file="comparison3.txt",append=TRUE,quote=FALSE,sep="\t",row.
>>>> na
>>>> mes=TRUE,col.names=FALSE)
>>>> 
>>>>   I then use the written out txt files (comparison1.txt,
>>>> comparison2.txt and comparison3.txt) to select significant probes
>>>> on the basis of log2fold change and adjusted p values thresholds, using
>>>> Excel.
>>>>   Would you say that this is a correct way to do this and could you
>>>> please recommend me some other, may be more efficient way of
>>>> writing the results of topTable for all 30000 probes out?
>>>> 
>>>>   With kind regards,
>>>>   Lev.
>>> 
>>> _______________________________________________
>>> 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
>> 
>> 
>> ___________________________________________________________________
>> The contents of this e-mail are privileged and/or confid...{{dropped:6}}
>> 
>> _______________________________________________
>> 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


___________________________________________________________________
The contents of this e-mail are privileged and/or confid...{{dropped:6}}



More information about the Bioconductor mailing list