[BioC] limma backgroundCorrect problem

[Gordon Smyth]

The normexp likelihood is difficult to compute with full numerical 
accuracy. In limma version 2.3.1, I made what I thought was a slight 
improvement in the likelihood calculation, which has instead had the 
effect of causing the normexp algorithm to sometimes (more often than 
before) converge to extreme parameter values, producing the effect 
that you see.

I believe the problem is fixed in limma 2.4.5 and later. Try upgrading and see.

Best wishes
Gordon

>Date: Thu, 26 Jan 2006 16:46:04 +0100
>From: Georg Otto <georg.otto at tuebingen.mpg.de>
>Subject: [BioC] limma backgroundCorrect problem
>To: bioconductor at stat.math.ethz.ch
>
>Hi,
>
>I am using limma for background correction and normalization of
>two-color arrays. I have encountered a problem that I do not quite
>understand: Using
>
> > RGb<-backgroundCorrect(RG, method="normexp", offset=50)
>
>I get the warning message:
>
>NaNs produced in: pnorm(q, mean, sd, lower.tail, log.p)
>
>
>However
>
> > which(is.nan(RGb$R) == TRUE)
> > which(is.nan(RGb$R) == TRUE)
>
>does not indicate NANs in the result. The numbers pruduced look
>strange as well:
>
> > RGb$G[1:5,]
>                            array1                       array2
>[1,]                 2.481014e+79                 2.361013e+74
>[2,]                 2.481014e+79                 2.361013e+74
>[3,]                 2.481014e+79                 2.361013e+74
>[4,]                 2.481014e+79                 2.361013e+74
>[5,]                 2.481014e+79                 2.361013e+74
>                            array3                       array4
>[1,]                   2099.87545                 1.577257e+79
>[2,]                     53.75734                 1.577257e+79
>[3,]                    118.30759                 1.577257e+79
>[4,]                     63.07401                 1.577257e+79
>[5,]                    339.25502                 1.577257e+79
>                           array5                      array6
>[1,]                1.153553e+85                  1644.20247
>[2,]                1.153553e+85                    51.08563
>[3,]                1.153553e+85                   138.33239
>[4,]                1.153553e+85                    62.48045
>[5,]                1.153553e+85                   612.04439
>
>
>My original foreground and background data however look decent (I
>think)
>
>
> > RG$G[1:5,]
>                            array1                       array2
>[1,]                   1827.41975                   1744.73457
>[2,]                     55.50345                     55.98621
>[3,]                    110.92857                    128.59155
>[4,]                     60.77931                     62.81119
>[5,]                     81.86429                    128.24161
>                            array3                       array4
>[1,]                   2102.72500                   1573.95541
>[2,]                     55.60690                     53.40936
>[3,]                    120.15714                     96.55072
>[4,]                     66.92357                     60.33103
>[5,]                    341.10458                    262.14966
>                           array5                      array6
>[1,]                  2279.76687                  1645.17073
>[2,]                    54.75172                    52.96226
>[3,]                   130.19728                   139.30065
>[4,]                    70.24615                    64.44872
>[5,]                   709.73718                   614.01266
>
>
> > RG$Gb[1:5,]
>                            array1                       array2
>[1,]                           44                           45
>[2,]                           44                           44
>[3,]                           44                           45
>[4,]                           45                           45
>[5,]                           44                           45
>                            array3                       array4
>[1,]                           48                           46
>[2,]                           47                           47
>[3,]                           47                           46
>[4,]                           49                           46
>[5,]                           47                           46
>                           array5                      array6
>[1,]                          47                          46
>[2,]                          47                          47
>[3,]                          46                          46
>[4,]                          47                          47
>[5,]                          46                          47
>
>
>Other subtraction methods (eg. "subtract") work well.
>
>I am running limma 2.4.4 on  R 2.2.1
>
>Any idea what is going wrong here?
>
>Best,
>
>Georg