Create a gist now

Instantly share code, notes, and snippets.

What would you like to do?
custom CDF in R / Bioconductor

Creating custom CDF for Affy chips in R / Bioconductor

knit_hooks$set(inline = identity) # sets output for inline resutls nicely
knit_hooks$set(error = function(x, options) stop(x)) # kills knitr if there
# is an error, therefore we don't waste time generating error messages
options(save.defaults=list(compress="xz"), stringsAsFactors=FALSE) 
# set up the default compression, and do NOT allow strings in data frames to become factors.
# That one is really important, and continually messes me up.
startTime <- Sys.time() # when did we start

What?

For those who don't know, CDF files are chip definition format files that define which probes belong to which probesets, and are necessary to use any of the standard summarization methods such as RMA, and others.

Why?

Because we can, and because custom definitions have been shown to be quite useful. See the information over at Brainarray.

Why not somewhere else?

A lot of times other people create custom CDF files based on their own criteria, and make it subsequently available for others to use (see the Brainarray for an example of what some are doing, as well as [PlandbAffy][linkplandb])

You have a really nifty idea for a way to reorganize the probesets on an Affymetrix chip to perform a custom analysis, but you don't want to go to the trouble of actually creating the CDF files and Bioconductor packages normally required to do the analysis, and yet you want to test and develop your analysis method.

How?

It turns out you are in luck. At least for AffyBatch objects in Bioconductor (created by calling ReadAffy), the CDF information is stored as an attached environment that can be easily hacked and modified to your hearts content. Environments in R are quite important and useful, and I wouldn't have come up with this if I hadn't been working in R for the past couple of years, but figured someone else might benefit from this knowledge.

The environment

In R, one can access an environment like so:

get("objName", envName) # get the value of object in the environment
ls(envName)

What is also very cool, is that one can extract the objects in an environment to a list, and also create their own environment from a list using list2env. Using this methodology, we can create our own definition of probesets that can be used by standard Bioconductor routines to summarize the probes into probesets.

A couple of disclaimers:

  • I have only tried this on 3' expression arrays
  • There might be a better way to do this, but I couldn't find it (let me know in the comments)

Example

require(affy)
require(estrogen)
require(hgu95av2cdf)

datadir = system.file("extdata", package="estrogen")

pd = read.AnnotatedDataFrame(file.path(datadir, "estrogen.txt"), header=TRUE, sep="", row.names=1)
pData(pd)

celDat = ReadAffy(filenames = rownames(pData(pd)), 
             phenoData = pd,
             verbose=TRUE, celfile.path=datadir)

This loads up the data, reads in the raw data, and gets it ready for us to use. Now, lets see what is in the actual CDF environment.

topProbes <- head(ls(hgu95av2cdf)) # get a list of probesets
topProbes

exSet <- get(topProbes[1], hgu95av2cdf)
exSet

We can see here that the first probe set r topProbes[1] has r nrow(exSet) perfect-match and mis-match probes in associated with it.

Lets summarize the original data using RMA.

rma1 <- exprs(rma(celDat))

head(rma1)

Now lets get the data as a list, and then create a new environment to be used for summarization.

allSets <- ls(hgu95av2cdf)
allSetDat <- mget(allSets, hgu95av2cdf)

allSetDat[1]

hgu2 <- list2env(allSetDat)
celDat@cdfName <- "hgu2"

rma2 <- exprs(rma(celDat))
head(rma2)

What about removing the MM columns? RMA only uses the PM, so it should still work.

allSetDat <- lapply(allSetDat, function(x){
	x[,1, drop=F]
})

allSetDat[1]

hgu3 <- list2env(allSetDat)
celDat@cdfName <- "hgu3"
rma3 <-exprs(rma(celDat))
head(rma3)

What if we only want to use the first 5 probesets?

allSetDat <- allSetDat[1:5]
hgu4 <- list2env(allSetDat)
celDat@cdfName <- "hgu4"
celDat
rma4 <- exprs(rma(celDat))
rma4
dim(rma4)

Custom CDF

To generate our custom CDF, we are going to set our own names, and take random probes from all of the probes on the chip. The actual criteria of which probes should be together can be made using any method the author chooses.

maxIndx <- 640*640

customCDF <- lapply(seq(1,100), function(x){
	tmp <- matrix(sample(maxIndx, 20), nrow=20, ncol=1)
	colnames(tmp) <- "pm"
	return(tmp)
})

names(customCDF) <- seq(1, 100)

hgu5 <- list2env(customCDF)
celDat@cdfName <- "hgu5"
rma5 <- exprs(rma(celDat))

head(rma5)

I hope this information is useful to someone else. I know it made my life a lot easier.

sessionInfo()

Edit: added session information.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment