[BioC] Re: limma plot error?

[Gordon Smyth]

At 03:46 PM 24/12/2003, Simon Melov wrote:
>when use the following series of commands to plot non-normalized data in 
>an MA plot highlighting the buffer controls and all other genes, I get a 
>very strange error. My feeling is this is occurring during the 
>normalization itself.

What you are describing seems impossible. limma does not do any adjustment 
of A-values anywhere unless you used normalizeBetweenArrays(). In 
particular I don't understand how you can attribute errors to the 
normalization step since your code and text indicate that you haven't done 
any normalization.

Perhaps you are plotting different things in GeneTraffic and limma. Have 
you background corrected in both? With the same background measure? Have 
you done any filtering of spots in GeneTraffic? Have you identified the 
same group of Buffer spots in both programs? If you use plotMA(RG) where do 
the Buffer spots appear?

Without seeing a reproducible example, it's not possible to say any more 
than this.

Gordon

>files <- dir(pattern="*.gpr")
>RG <- read.maimages(files,source="genepix")
>targets <- readTargets("targets.txt")
>RG$genes <- readGAL()
>RG$printer <- getLayout(RG$genes)
>targets <- readTargets("targets.txt")
>types <- readSpotTypes("Spottypes.txt")
>RG$genes$Status <- controlStatus(types, RG)
>
>Matching patterns for: ID Name
>Found 25088 gene
>Found 32 Calibration-1
>Found 32 Calibration-2
>Found 32 Calibration-3
>Found 32 Calibration-4
>Found 32 Calibration-5
>Found 32 Calibration-6
>Found 32 Calibration-7
>Found 32 Calibration-8
>Found 32 Calibration-9
>Found 32 Calibration-10
>Found 512 Empty
>Found 288 Buffer
>Found 0 No data
>Found 32 Human Cot
>Found 32 Mouse Cot
>Found 32 B Actin
>Found 32 Poly A
>Found 32 Salmon Sperm
>Setting attributes: values Color
>
>plotMA(RG,value=c("gene","Buffer"),col=c("black","blue"))
>
>In the input data I have checked for correct assignation of block, column, 
>row (I am using gpr files) and gene identity. After normalization my 
>buffers (which are negative controls) are displayed across the range of A 
>values in the MA plot (as seen below). However, most of my buffers have 
>very low A values (if you calculate manually). Somehow, between loading 
>and calculation of the A values, limma is adjusting these buffer values so 
>that they are spread across the range. I have checked everything I can 
>(i.e. are the individual values (R, Rb etc) being pulled from the correct 
>columns in the gpr files, and being assigned to the correct position in 
>the matrix. They are.
>
>As a final check, I plotted the same data in Genetraffic, and then in 
>limma (both non-normalized), with the buffers highlighted in both plots. 
>Its the same data, but you can see when the raw data is plotted in limma 
>versus in Genetraffic, it is somehow being plotted incorrectly. There are 
>288 buffer spots being highlighted in blue in each graph, but as you can 
>see, the values in the limma graph are all spread out. This seems to be 
>occurring during the normalization function, as the values going into the 
>normalization function are correct, but those coming out are wrong. 
>Individually checking several of the buffer values confirms this. I am not 
>clear how this is occurring. Unless I am doing something very 
>fundamentally wrong, then this is a big problem.
>
>Please help!
>
>thanks
>
>Simon.
>
>Blue are the buffer controls in both cases. 256 of the buffer values are 
>below an A value of 7. Only 8 values are above 8. But in the limma plot, 
>most buffers are being plotted with an A value of greater than 8, which is 
>clearly fundamentally different.