[BioC] Inf when using duplicateCorrelation
Gordon K Smyth
smyth at wehi.EDU.AU
Sat Jan 24 03:34:45 CET 2009
Dear Ingrid,
You cannot use a blocking variable which is entirely confounded with the
design matrix. In your experiment, there is no overlap in treatment
groups between the first and second blocks, hence there is no block effect
remaining to estimate after the design matrix is fitted. Hence the
estimated correlations are degenerate.
Best wishes
Gordon
> Date: Thu, 22 Jan 2009 16:28:31 +0100
> From: Ingrid H. G. ?stensen <Ingrid.H.G.Ostensen at rr-research.no>
> Subject: [BioC] Inf when using duplicateCorrelation
> To: <bioconductor at stat.math.ethz.ch>
>
> Hi
>
> I am trying to analyze a data set consisting of data run on two
> different times a few months a part. The data set consists of 8 groups
> with 3 biological replicates in each, and Illumina Human WG6 v3 arrays
> have been used. I am using the probe profile file in the analysis.
>
> After the quality control it looks like the data is separated into the
> different groups (8), but I can also slightly see the arrays separate
> them self into the two groups based on when they were run.
>
> To try to block the effect caused by the two lab periods I thought of
> using duplicateCorrelation. Unfortunately I can not get it to work this
> time,
>
> This is my design matrix:
>> designMa
> S0_s S18_s S1_s S4_s T0_s T18_s T1_s T4_s
> S_0h 1 0 0 0 0 0 0 0
> S_0h 1 0 0 0 0 0 0 0
> S_0h 1 0 0 0 0 0 0 0
> T_0h 0 0 0 0 1 0 0 0
> T_0h 0 0 0 0 1 0 0 0
> T_0h 0 0 0 0 1 0 0 0
> S_1h 0 0 1 0 0 0 0 0
> S_1h 0 0 1 0 0 0 0 0
> S_1h 0 0 1 0 0 0 0 0
> T_1h 0 0 0 0 0 0 1 0
> T_1h 0 0 0 0 0 0 1 0
> T_1h 0 0 0 0 0 0 1 0
> S_4h 0 0 0 1 0 0 0 0
> S_4h 0 0 0 1 0 0 0 0
> S_4h 0 0 0 1 0 0 0 0
> T_4h 0 0 0 0 0 0 0 1
> T_4h 0 0 0 0 0 0 0 1
> T_4h 0 0 0 0 0 0 0 1
> S_18h 0 1 0 0 0 0 0 0
> S_18h 0 1 0 0 0 0 0 0
> S_18h 0 1 0 0 0 0 0 0
> T_18h 0 0 0 0 0 1 0 0
> T_18h 0 0 0 0 0 1 0 0
> T_18h 0 0 0 0 0 1 0 0
>
> S0, T0 and S1 are in the first run and the rest in the second.
>
> dataSet_Norm_exp_log2_ordnet is my normalized expression data as a matrix and blokk looks like this:
>> blokk
> [1] 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2
>
>> corfit <- duplicateCorrelation(dataSet_Norm_exp_log2_ordnet, design = designMa, ndups = 1, block = as.factor(blokk))
> 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
>
>
>> fitDesMa <- lmFit(dataSet_Norm_exp_log2_ordnet,design = designMa,block = as.factor(blokk),cor = corfit$consensus)
> Error in chol.default(V) : the leading minor of order 2 is not positive definite
>
>> corfit
> $consensus.correlation
> [1] 1
>
> $cor
> [1] 1
>
> $atanh.correlations
> [1] Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf
> [39] Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf
> [77] Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf
> [115] Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf
> [153] Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf
> [191] Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf Inf
>
> Does any one have any suggestions for why I get all the Inf? Maybe
> duplicateCorrelation is not the best thing?
>
> Regards,
> Ingrid
More information about the Bioconductor
mailing list