[BioC] Single Channel Analysis in Limma using lmscFit

[Gordon Smyth]

>Date: Mon, 11 Apr 2005 13:08:55 -0700
>From: Brett Abrahams <bsa at ucla.edu>
>Subject: [BioC] Single Channel Analysis in Limma using lmscFit
>To: bioconductor at stat.math.ethz.ch
>
>Hello,
>
>I would like to use limma to carry out single channel analyses on some two
>color data (to address comparisons that can't be made with standard methods
>and my unconnected experimental design). I'm able to get two color analyses
>to work nicely but run into problems when I try to run the single channel
>analysis using the 'lmscFit' function (as described in the March 9 2005
>user's guide) .
>
>Everything seems to work fine until I run 'intraspotCorrelation' which
>results in a number of errors (see Point 1 below for input/output).

These are not "errors". They are warnings.

>  I've
>looked into the versions of limma (Version: 1.8.22) and statmod (Version:
>1.1.0) but this doesn't seem to be the answer as both are current. Any
>thoughts on why these error messages are being generated and what I can do
>to fix the problem would be much appreciated. Also, am I right in thinking
>the 'reml' errors I get refer to problems with only single genes? I picked
>this up from a previous post to the list but may have misinterpreted.
>
>If I ignore the errors and carry on with the analysis through to the
>topTable I get more results that I can't understand (see Point 2 below
>input/output). What's confusing me here is that although results from the
>'decideTests' function seems to suggest that differentially expressed genes
>are present within each of the four contrasts I've specified, only one of
>the four corresponding topTables shows anything with significant p values.
>Amongst the contrasts without significant differences most p values
>are >0.5 and all B values are negative. Any clarification would be great.

The help page for 'decideTests' says

    "The default settings with 'method="separate"' is equivalent to using 
'topTable'"

But you have specified the different 'method="nestedF".

Gordon

>Thanks in advance for this wonderful software and superb documentation /
>support.
>
>Bret
>
>Point 1
>  > corfit <- intraspotCorrelation(MA, design)
>
>Loading required package: statmod
>
>Attaching package 'statmod':
>
>The following object(s) are masked from package:limma :
>
>matvec vecmat
>
>Warning messages:
>
>1: reml: Max iterations exceeded in: remlscore(y, X, Z)
>2: reml: Max iterations exceeded in: remlscore(y, X, Z)
>3: reml: Max iterations exceeded in: remlscore(y, X, Z)
>
>  >
>
>Point 2
>
>Two groups (G1 and G2) with two tissues examined for each
>
>  > results <- decideTests(fit, method="nestedF")
>
>  > summary(results)
>g1-g2   t1-t2   g1t2 - g2t2     g1t1-g2t1
>-1    29   936      30          14
>0  17866 16356   17861       17891
>1     60   663      64          50
>
>  > topTable(fit3, coef=1, adjust="fdr")
>  > topTable(fit3, coef=1, adjust="fdr")
>
>   Status          M        A         t   P.Value         B
>15540   gene  0.4363826 8.514271  5.138324 0.6918086 -1.734635
>10050   gene -0.3957220 9.904263 -4.802365 0.6918086 -1.936940
>13504   gene  0.4280439 8.957909  4.492316 0.6918086 -2.136429
>9953    gene  0.3380621 8.555516  4.486686 0.6918086 -2.140164
>3266    gene  0.4737100 6.280065  4.478957 0.6918086 -2.145299
>5596    gene  0.3738408 8.360172  4.327564 0.6918086 -2.247386
>18159   gene -0.3860020 6.880403 -4.272878 0.6918086 -2.284962
>6550    gene  0.4545545 7.901485  4.258022 0.6918086 -2.295233
>8589    gene -0.4707301 7.850181 -4.227162 0.6918086 -2.316657
>10149   gene  0.3387359 9.005838  4.216223 0.6918086 -2.324278