[BioC] Problem with lmfit

[Gordon K Smyth]

Dear David,

limma doesn't work with only one gene.  In principle, it should give the 
same results as lm when there is only one row of data.  In practice, it 
fails for a purely technical programming issue -- R by default converts 
single row matrices into vectors, and this confuses limma which is 
expecting to get matrices.  We could fix this, but no one has ever asked 
for it before!

Best wishes
Gordon


> Date: Mon, 09 Aug 2010 15:50:33 +0200
> From: David martin <vilanew at gmail.com>
> To: bioconductor at stat.math.ethz.ch
> Subject: Re: [BioC] Problem with lmfit
> Message-ID: <i3p14l$unk$2 at dough.gmane.org>
> Content-Type: text/plain; charset=ISO-8859-1; format=flowed
>
> Thanks Paolo,
> You are indeed right, limma is behaving strange with only 1 feature. The
> thing is that i'm testing a single gene over a large panel of samples
> (this is only a subset)..
> thanks for letting me know
>
> On 09/08/10 15:29, Paolo Innocenti wrote:
>> Hi,
>>
>> I can reproduce the error, which seems to be caused by your
>> expressionset having only one feature. If you try with 2 lines, it works:
>>
>> library(Biobase)
>> library(limma)
>> design <- matrix(c(1,0),ncol=2,nrow=5,byrow=T)
>> design <- rbind(design,c(0,1),c(0,1))
>> colnames(design) <- c("HEALTHY","DISEASE")
>>
>> # 2 (identical) features
>>
>> data <- matrix(rep(c(25.92000, 25.66667, 23.14667, 26.28667, 25.48667
>> ,21.19000, 20.81000),2),nrow=2,ncol=7, byrow=T)
>> mydata.eset=new("ExpressionSet",exprs=data)
>> lmFit(mydata.eset, design=design)
>>
>> I have no idea if lmFit is meant to work with 1 feature (I guess it
>> yield the same result as "lm"), but the error seem to be caused by the
>> function "nonEstimable" called by limma (but I am no expert, someone
>> else might want to comment on this).
>>
>> Hope this helps.
>> paolo
>>
>> # lm code:
>> y <- c(25.92000, 25.66667, 23.14667, 26.28667, 25.48667 ,21.19000,
>> 20.81000)
>> treat <- factor(c(rep("healthy",5),rep("dis",2)))
>> summary(lm(y ~ treat))
>>
>>
>> On 08/09/2010 02:59 PM, David martin wrote:
>>> Hello,
>>> I have a strange problem while fitting a design to my data. Here is the
>>> output code. You can reproduce the problem by copy paste the code below.
>>>
>>>
>>>> design
>>> HEALTHY DISEASE
>>> 1 1 0
>>> 2 1 0
>>> 3 1 0
>>> 4 1 0
>>> 5 1 0
>>> 6 0 1
>>> 7 0 1
>>> attr(,"assign")
>>> [1] 1 1
>>> attr(,"contrasts")
>>> attr(,"contrasts")$groups
>>> [1] "contr.treatment"
>>>
>>>> exprs(mydata.eset)
>>> healthy1 healthy2 healthy3 healthy4 N25 disease1 disease2
>>> 1 25.92 25.66667 23.14667 26.28667 25.48667 21.19 20.81
>>>
>>>> lmFit(exprs(mydata.eset), design=design)
>>> Error in fit$effects[(fit$rank + 1):narrays, , drop = FALSE] :
>>> incorrect number of dimensions
>>>
>>>
>>> ### Here is the code to reproduce the error
>>> library(Biobase)
>>> library(limma)
>>> design <- matrix(c(1,0),ncol=2,nrow=5,byrow=T)
>>> design <- rbind(design,c(0,1),c(0,1))
>>> colnames(design) <- c("HEALTHY","DISEASE")
>>> data <- matrix(c(25.92000, 25.66667, 23.14667, 26.28667, 25.48667
>>> ,21.19000, 20.81000),nrow=1,ncol=7)
>>> mydata.eset=new("ExpressionSet",exprs=data)
>>> colnames(exprs(mydata.eset)) <-
>>> c("healthy1","healthy2","healthy3","healthy4","healthy5","disease1","disease2")
>>>
>>>
>>> lmFit(mydata.eset, design=design)
>>>
>>>
>>> thanks for leting me know,

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