In Bioconductor/Biostrings: Efficient manipulation of biological strings

library(BiocStyle)

Overview

This document presents some basic and simple tools for dealing with the oligonucleotide microarray reporter sequence information in the Bioconductor probe packages. This information is used, for example, in the r Biocpkg("gcrma") package.

Probe packages are a convenient way for distributing and storing the probe sequences (and related information) of a given chip.

As an example, the package r Biocpkg("hgu95av2probe") provides microarray reporter sequences for Affymetrix' HgU95a version 2 genechip, and for almost all major Affymetrix genechips, the corresponding packages can be downloaded from the Bioconductor website. If you have the reporter sequence information of a particular chip, you can also create such a package yourself. This is described in the makeProbePackage vignette of the r Biocpkg("AnnotationForge") package.

This document assumes some basic familiarity with R and with the design of the AffyBatch class in the r Biocpkg("affy") package, Bioconductor's basic container for Affymetrix genechip data.

First, let us load the r Biocpkg("Biostrings") package and some other packages we will use.

library(Biostrings) 
library(hgu95av2probe) 
library(hgu95av2cdf)

Using probe packages

Help for the probe sequence data packages can be accessed through

?hgu95av2probe

One of the issues that you have to deal with is that the probe packages do not provide the reporter sequences of all the features present in an AffyBatch. Some sequences are missing, some are implied; in particular, the data structure in the probe packages does not explicitly contain the sequences of the mismatch probes, since they are implied by the perfect match probes. Also, some other features, typically harboring control probes or empty, do not have sequences. This is the choice that Affymetrix made when they made files with probe sequences available, and we followed it.

Practically, this means that the vector of probe sequences in a probe package does not align 1:1 with the rows of the corresponding AffyBatch; you need to keep track of this mapping, and some tools for this are provided and explained below (see Section 2.2). It also means that some functions from the r Biocpkg("affy") package, such as pm, cannot be used when the sequences of the probes corresponding to their result are needed; since pm reports the intensities, but not the identity of the probes it has selected, yet the latter would be needed to retrieve their sequences.

Basic functions

Let us look at some basic operations on the sequences.

Reverse and complementary sequence

DNA sequences can be reversed and/or complemented with the reverse, complement and reverseComplement functions.

?reverseComplement

Matching sets of probes against each other

pm <- DNAStringSet(hgu95av2probe) 
dict <- pm[3801:4000]
pdict <- PDict(dict)
m <- vcountPDict(pdict, pm)
dim(m) 
table(rowSums(m))
which(rowSums(m) == 3)
ii <- which(m[77, ] != 0)
pm[ii]

Base content

The base content (number of occurrence of each character) of the sequences can be computed with the function alphabetFrequency.

bcpm <- alphabetFrequency(pm, baseOnly=TRUE)
head(bcpm) 
alphabetFrequency(pm, baseOnly=TRUE, collapse=TRUE)

Relating to the features of an AffyBatch {#subsec.relating}

nc = hgu95av2dim$NCOL
nc

nr = hgu95av2dim$NROW
nr

Each column of an AffyBatch corresponds to an array, each row to a certain probe on the arrays. The probes are stored in a way that is related to their geometrical position on the array. For example, the hgu95av2 array is geometrically arranged into r nc columns and r nr rows. We call the column and row indices the x- and y-coordinates, respectively. This results in r nc ⨉ r nr = r format(nc*nr, scientific=FALSE) probes of the AffyBatch; we also call them indices. To convert between x-and y-coordinates and indices, you can use the functions xy2indices and indices2xy from the r Biocpkg("affy") package.

The sequence data in the probe packages is addressed by their x and y-coordinates. Let us construct a vector abseq that aligns with the indices of an hgu95av2 AffyBatch and fill in the sequences:

library(affy) 
abseq = rep(as.character(NA), nc*nr) 
ipm = with(hgu95av2probe, xy2indices(x, y, nc=nc)) 
any(duplicated(ipm)) # just a sanity check 
abseq[ipm] = hgu95av2probe$sequence
table(is.na(abseq))

The mismatch sequences are not explicitly stored in the probe packages. They are implied by the match sequences, by flipping the middle base. This can be done with the pm2mm function defined below. For Affymetrix GeneChips the length of the probe sequences is 25, and since we start counting at 1, the middle position is 13.

mm <- pm
subseq(mm, start=13, width=1) <- complement(subseq(mm, start=13, width=1))
cat(as.character(pm[[1]]), as.character(mm[[1]]), sep="\n")

We compute imm, the indices of the mismatch probes, by noting that each mismatch has the same x-coordinate as its associated perfect match, while its y-coordinate is increased by 1.

imm = with(hgu95av2probe, xy2indices(x, y+1, nc=nc))
intersect(ipm, imm) # just a sanity check
abseq[imm] = as.character(mm)
table(is.na(abseq))

See Figures \@ref(fig:bap)-\@ref(fig:p2p) for some applications of the probe sequence information to preprocessing and data quality related plots.

Some sequence related "preprocessing and quality" plots

The function alphabetFrequency counts the number of occurrences of each of the four bases A, C, G, T in each probe sequence.

freqs <- alphabetFrequency(DNAStringSet(abseq[!is.na(abseq)]), baseOnly=TRUE)
bc <- matrix(nrow=length(abseq), ncol=5)
colnames(bc) <- colnames(freqs)
bc[!is.na(abseq), ] <- freqs
head(na.omit(bc))

Let us define an ordered factor variable for GC content:

GC = ordered(bc[,"G"] + bc[,"C"])
colores = rainbow(nlevels(GC))

And let us create an AffyBatch object.

library(affydata)
f <- system.file("extracelfiles", "CL2001032020AA.cel", package="affydata")
pd <- new("AnnotatedDataFrame", data=data.frame(fromFile=I(f), row.names="f"))
abatch <- read.affybatch(filenames=f, compress=TRUE, phenoData=pd)
barplot(table(GC), col=colores, xlab="GC", ylab="number")

boxplot(log2(exprs(abatch)[,1]) ~ GC, outline=FALSE,
        col=colores, , xlab="GC", ylab=expression(log[2]~intensity))

plot(exprs(abatch)[ipm,1], exprs(abatch)[imm,1], asp=1, pch=".", log="xy",
     xlab="PM", ylab="MM", col=colores[GC[ipm]])
abline(a=0, b=1, col="#404040", lty=3)