[BioC] limma, duplicateCorrelation, dupfit$consensus = 1

[Gordon K Smyth]

Dear Karl,

This result suggests that you should average your replicate spots (see the avedups function in
limma) rather than using duplicate correlation.  Yes, it almost certainly results from trying to
do too much with too little data.  Either you don't have enough data, or the model you're fitting
contains too many terms, so that you can't get a reliable estimate of the within-array
correlation.

Best wishes
Gordon

> Date: Mon, 24 May 2010 16:01:11 +0200
> From: Karl Brand <k.brand at erasmusmc.nl>
> To: "bioconductor at stat.math.ethz.ch" <bioconductor at stat.math.ethz.ch>
> Subject: Re: [BioC] limma, duplicateCorrelation,	dupfit$consensus = 1:
> 	concerning?
>
> I see now that i can not proceed with such a design including
> duplicateCorrelation (output below).  Perhaps it results from trying to
> do to much with too few observations/replicates.
>
> I'd still really like to hear explanations why this might be so.
>
> cheers,
>
> Karl
>
>  >      fit <- lmFit(rma.pp, design, correlation=dupfit$consensus,
> block=Animal)
> Error in chol.default(V) :
>    the leading minor of order 2 is not positive definite
>  >
>
> On 5/24/2010 3:18 PM, Karl Brand wrote:
>> Dear BioC,
>>
>> I'm attempting to use the duplicateCorrelation function within limma to
>> control for fact that tissues being studied come from the same animals.
>> I understand that for this work appropriately, dupfit$consensus needs to
>> be positive, which at "1", it is.
>>
>> But such an exact value is concerning to my inexperienced eye.
>>
>> I'd greatly appreciate hearing from people with experience using
>> duplicateCorrelation what this outcome represents and if indeed is
>> something to be wary of and perhaps omitted.
>>
>> With thanks in advance,
>>
>> cheers,
>>
>> Karl
>>
>>> design <- model.matrix(~Time * Pperiod * Tissue - Time : Pperiod :
>> Tissue)
>>> source(.trPaths[5], echo=TRUE, max.deparse.length=150)
>>
>>> dupfit <- duplicateCorrelation(rma.pp, design, ndups=1, block=Animal)
>>
>>> dupfit$consensus #remembering to check that dupfit$consensus is
>> positive.
>> [1] 1
>> There were 50 or more warnings (use warnings() to see the first 50)
>>> warnings()
>> Warning messages:
>> 1: In sqrt(dfitted.values) : NaNs produced
>> 2: In sqrt(dfitted.values) : NaNs produced
>> 3: In sqrt(dfitted.values) : NaNs produced
>> 4: In sqrt(dfitted.values) : NaNs produced
>> 5: In sqrt(dfitted.values) : NaNs produced
>> 6: In sqrt(dfitted.values) : NaNs produced
>> 7: In sqrt(dfitted.values) : NaNs produced
>> 8: In sqrt(dfitted.values) : NaNs produced
>> 9: In sqrt(dfitted.values) : NaNs produced
>> 10: In sqrt(dfitted.values) : NaNs produced
>> 11: In sqrt(dfitted.values) : NaNs produced
>> 12: In sqrt(dfitted.values) : NaNs produced
>> 13: In sqrt(dfitted.values) : NaNs produced
>> 14: In sqrt(dfitted.values) : NaNs produced
>> 15: In sqrt(dfitted.values) : NaNs produced
>> 16: In sqrt(dfitted.values) : NaNs produced
>> 17: In sqrt(dfitted.values) : NaNs produced
>> 18: In sqrt(dfitted.values) : NaNs produced
>> 19: In sqrt(dfitted.values) : NaNs produced
>> 20: In sqrt(dfitted.values) : NaNs produced
>> 21: In sqrt(dfitted.values) : NaNs produced
>> 22: In sqrt(dfitted.values) : NaNs produced
>> 23: In sqrt(dfitted.values) : NaNs produced
>> 24: In sqrt(dfitted.values) : NaNs produced
>> 25: In sqrt(dfitted.values) : NaNs produced
>> 26: In sqrt(dfitted.values) : NaNs produced
>> 27: In sqrt(dfitted.values) : NaNs produced
>> 28: In sqrt(dfitted.values) : NaNs produced
>> 29: In sqrt(dfitted.values) : NaNs produced
>> 30: In sqrt(dfitted.values) : NaNs produced
>> 31: In sqrt(dfitted.values) : NaNs produced
>> 32: In sqrt(dfitted.values) : NaNs produced
>> 33: In sqrt(dfitted.values) : NaNs produced
>> 34: In sqrt(dfitted.values) : NaNs produced
>> 35: In sqrt(dfitted.values) : NaNs produced
>> 36: In sqrt(dfitted.values) : NaNs produced
>> 37: In sqrt(dfitted.values) : NaNs produced
>> 38: In sqrt(dfitted.values) : NaNs produced
>> 39: In sqrt(dfitted.values) : NaNs produced
>> 40: In sqrt(dfitted.values) : NaNs produced
>> 41: In sqrt(dfitted.values) : NaNs produced
>> 42: In sqrt(dfitted.values) : NaNs produced
>> 43: In sqrt(dfitted.values) : NaNs produced
>> 44: In sqrt(dfitted.values) : NaNs produced
>> 45: In sqrt(dfitted.values) : NaNs produced
>> 46: In sqrt(dfitted.values) : NaNs produced
>> 47: In sqrt(dfitted.values) : NaNs produced
>> 48: In sqrt(dfitted.values) : NaNs produced
>> 49: In sqrt(dfitted.values) : NaNs produced
>> 50: In sqrt(dfitted.values) : NaNs produced
>>>
>>
>>
>>
>
> --
> Karl Brand
> Department of Genetics
> Erasmus MC
> Dr Molewaterplein 50
> 3015 GE Rotterdam
> T +31 (0)10 704 3457 | F +31 (0)10 704 4743 | M +31 (0)642 777 268



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