lmFit: Linear Model for Series of Arrays

Description

Fit linear model for each gene given a series of arrays

Usage

lmFit(object, design=NULL, ndups=1, spacing=1, block=NULL, correlation, weights=NULL, method="ls", ...)

Arguments

object

A matrix-like data object containing log-ratios or log-expression values for a series of arrays, with rows corresponding to genes and columns to samples. Any type of data object that can be processed by getEAWP is acceptable.

design

the design matrix of the microarray experiment, with rows corresponding to arrays and columns to coefficients to be estimated. Defaults to the unit vector meaning that the arrays are treated as replicates.

ndups

positive integer giving the number of times each distinct probe is printed on each array.

spacing

positive integer giving the spacing between duplicate occurrences of the same probe, spacing=1 for consecutive rows.

block

vector or factor specifying a blocking variable on the arrays. Has length equal to the number of arrays. Must be NULL if ndups>2.

correlation

the inter-duplicate or inter-technical replicate correlation

weights

non-negative observation weights. Can be a numeric matrix of individual weights, of same size as the object expression matrix, or a numeric vector of array weights with length equal to ncol of the expression matrix, or a numeric vector of gene weights with length equal to nrow of the expression matrix.

method

fitting method; "ls" for least squares or "robust" for robust regression

...

other optional arguments to be passed to lm.series, gls.series or mrlm

Value

An MArrayLM object containing the result of the fits.The rownames of object are preserved in the fit object and can be retrieved by rownames(fit) where fit is output from lmFit. The column names of design are preserved as column names and can be retrieved by colnames(fit).

Details

This function fits multiple linear models by weighted or generalized least squares. It accepts data from a experiment involving a series of microarrays with the same set of probes. A linear model is fitted to the expression data for each probe. The expression data should be log-ratios for two-color array platforms or log-expression values for one-channel platforms. (To fit linear models to the individual channels of two-color array data, see lmscFit.) The coefficients of the fitted models describe the differences between the RNA sources hybridized to the arrays. The probe-wise fitted model results are stored in a compact form suitable for further processing by other functions in the limma package.

The function allows for missing values and accepts quantitative weights through the weights argument. It also supports two different correlation structures. If block is not NULL then different arrays are assumed to be correlated. If block is NULL and ndups is greater than one then replicate spots on the same array are assumed to be correlated. It is not possible at this time to fit models with both a block structure and a duplicate-spot correlation structure simultaneously.

If object is a matrix then it should contain log-ratios or log-expression data with rows corresponding to probes and columns to arrays. (A numeric vector is treated the same as a matrix with one column.) For objects of other classes, a matrix of expression values is taken from the appropriate component or slot of the object. If object is of class MAList or marrayNorm, then the matrix of log-ratios (M-values) is extracted. If object is of class ExpressionSet, then the expression matrix is extracted. (This may contain log-expression or log-ratio values, depending on the platform.) If object is of class PLMset then the matrix of chip coefficients chip.coefs is extracted.

The arguments design, ndups, spacing and weights will be extracted from the data object if available and do not normally need to set explicitly in the call. On the other hand, if any of these are set in the function call then they will over-ride the slots or components in the data object. If object is an PLMset, then weights are computed as 1/pmax(object@se.chip.coefs, 1e-05)^2. If object is an ExpressionSet object, then weights are not computed.

If the argument block is used, then it is assumed that ndups=1.

The correlation argument has a default value of 0.75, but in normal use this default value should not be relied on and the correlation value should be estimated using the function duplicateCorrelation. The default value is likely to be too high in particular if used with the block argument.

The actual linear model computations are done by passing the data to one the lower-level functions lm.series, gls.series or mrlm. The function mrlm is used if method="robust". If method="ls", then gls.series is used if a correlation structure has been specified, i.e., if ndups>1 or block is non-null and correlation is different from zero. If method="ls" and there is no correlation structure, lm.series is used.

Examples

Run this code

# Simulate gene expression data for 100 probes and 6 microarrays
# Microarray are in two groups
# First two probes are differentially expressed in second group
# Std deviations vary between genes with prior df=4
sd <- 0.3*sqrt(4/rchisq(100,df=4))
y <- matrix(rnorm(100*6,sd=sd),100,6)
rownames(y) <- paste("Gene",1:100)
y[1:2,4:6] <- y[1:2,4:6] + 2
design <- cbind(Grp1=1,Grp2vs1=c(0,0,0,1,1,1))
options(digits=3)

# Ordinary fit
fit <- lmFit(y,design)
fit <- eBayes(fit)
topTable(fit,coef=2)
dim(fit)
colnames(fit)
rownames(fit)[1:10]
names(fit)

# Fold-change thresholding
fit2 <- treat(fit,lfc=0.1)
topTreat(fit2,coef=2)

# Volcano plot
volcanoplot(fit,coef=2,highlight=2)

# Mean-difference plot
plotMD(fit,column=2)

# Q-Q plot of moderated t-statistics
qqt(fit$t[,2],df=fit$df.residual+fit$df.prior)
abline(0,1)

# Various ways of writing results to file
## Not run: write.fit(fit,file="exampleresults.txt")
## Not run: write.table(fit,file="exampleresults2.txt")

# Fit with correlated arrays
# Suppose each pair of arrays is a block
block <- c(1,1,2,2,3,3)
dupcor <- duplicateCorrelation(y,design,block=block)
dupcor$consensus.correlation
fit3 <- lmFit(y,design,block=block,correlation=dupcor$consensus)

# Fit with duplicate probes
# Suppose two side-by-side duplicates of each gene
rownames(y) <- paste("Gene",rep(1:50,each=2))
dupcor <- duplicateCorrelation(y,design,ndups=2)
dupcor$consensus.correlation
fit4 <- lmFit(y,design,ndups=2,correlation=dupcor$consensus)
dim(fit4)
fit4 <- eBayes(fit4)
topTable(fit4,coef=2)