[BioC] warning messages from duplicateCorrelation in limma

[Na, Ren]

Dr. Smyth,

Thank you very much for your prompt reply. I tried remove missing and control spots exactly like the way you suggested before I sent my first email, and I got several same warnings too. But your explain clarified my thoughts and make me feel more comfortable with my data. Thanks again.

Ren 


-----Original Message-----
From: Gordon K Smyth [mailto:smyth at wehi.EDU.AU]
Sent: Thu 11/4/2004 5:41 AM
To: Na, Ren
Cc: bioconductor at stat.math.ethz.ch
Subject: [BioC] warning messages from duplicateCorrelation in limma
 
> Date: Wed, 3 Nov 2004 17:14:18 -0600
> From: "Na, Ren" <Na at uthscsa.edu>
> Subject: [BioC] warning messages from duplicateCorrelation in limma
> To: <bioconductor at stat.math.ethz.ch>
> Message-ID:
> 	<34263DF081CECE4F81FEF001C113CFA50206EA14 at addax.win.uthscsa.edu>
> Content-Type: text/plain
>
> Hi
> I have an another question about function duplicateCorrelation for my another data set which is
> four pairs of dye swap, four wild type biological replicates and four mutant biological
> replicates. When I run this function, I got Warning messages. I tried in two ways,

Firstly, warnings from duplicateCorrelation() are not a problem.  The warnings affect only 4 genes
out of many thousands.  The consensus correlation will be little affected if at all because it is
a robust average over all the genes.

Secondly, assigning zero weight is not the way to remove missing or control spots.  You should
leave the weights as they are are and instead subject the data object.  For example,

  isBlank <- (MA$genes$Status %in% c("blank","miss"))
  corfit <- duplicateCorrelation(MA[!isBlank,], design, ndups=1, block=pair)

See the Weaver example in the Limma User's Guide for an example of the treatment of control spots.

Gordon

> 1) I only assigned weight 0 to missing spots which were not printed on our slides
> by doing
> w<-modifyWeights(RG$weights,status=RG$genes$Status, "miss",0)
> RG$weights<-w
> and when I run
>>design<-cbind(Dye=1, XpavsWt=c(1,-1,1,-1,1,-1,1,-1))
>>pair<-c(1,1,2,2,3,3,4,4)
>> corfit<-duplicateCorrelation(MA,design,ndups=1,block=pair)
> Warning messages:
> 1: Too much damping - convergence tolerance not achievable in: glmgam.fit(dx, dy, start = start,
> tol = tol, maxit = maxit, trace = trace)
> 2: Too much damping - convergence tolerance not achievable in: glmgam.fit(dx, dy, start = start,
> tol = tol, maxit = maxit, trace = trace)
> 3: Too much damping - convergence tolerance not achievable in: glmgam.fit(dx, dy, start = start,
> tol = tol, maxit = maxit, trace = trace)
> 4: Too much damping - convergence tolerance not achievable in: glmgam.fit(dx, dy, start = start,
> tol = tol, maxit = maxit, trace = trace)
>
> corfit$consensus is -0.28
>
> 2) I assigned weight 0 to missing spots and blank spots on our slides like
> w<-modifyWeights(RG$weights,status=RG$genes$Status, c("blank","miss"),c(0,0))
> RG$weights<-w
> and when I run
>> corfit<-duplicateCorrelation(MA,design,ndups=1,block=pair)
> Warning messages:
> 1: Too much damping - convergence tolerance not achievable in: glmgam.fit(dx, dy, start = start,
> tol = tol, maxit = maxit, trace = trace)
>
> corfit$consensus is -0.31
>
> Blank spots are the spot which were printed with buffer.
> In second way, corfit$consensus is increased a little bit and I got one warning in stead of four
> warnings.
>
> My questions are: what does the warning message mean? Does it mean the arrays' reproducibility is
> poor? Can I still get reasonable top list of genes by using the second way? I would appreciate any
> suggestion.
>
> Thanks in advance,
> Ren





	[[alternative HTML version deleted]]