varFit | R Documentation |
Fit linear model on mean absolute or squared deviations for each CpG given a series of methylation arrays
varFit(
data,
design = NULL,
coef = NULL,
type = NULL,
trend = TRUE,
robust = TRUE,
weights = NULL
)
## S3 method for class 'MethylSet'
varFit(
data,
design = NULL,
coef = NULL,
type = NULL,
trend = TRUE,
robust = TRUE,
weights = NULL
)
## S3 method for class 'DGEList'
varFit(
data,
design = NULL,
coef = NULL,
type = NULL,
trend = TRUE,
robust = TRUE,
weights = NULL
)
## Default S3 method:
varFit(
data,
design = NULL,
coef = NULL,
type = NULL,
trend = TRUE,
robust = TRUE,
weights = NULL
)
data |
Object of class |
design |
The design matrix of the experiment, with rows corresponding to arrays/samples and columns to coefficients to be estimated. Defaults to the unit vector. |
coef |
The columns of the design matrix containing the comparisons to test for differential variability. Defaults to all columns of design matrix. |
type |
Character string, |
trend |
Logical, if true fits a mean variance trend on the absolute or squared deviations. |
robust |
Logical, if true performs robust empirical Bayes shrinkage of the variances for the moderated t statistics. |
weights |
Non-negative observation weights. Can be a numeric matrix of individual weights, of same size as the object matrix, or a numeric vector of array weights, or a numeric vector of gene/feature weights. |
This function depends on the limma
package and is used to rank
features such as CpG sites or genes in order of evidence of differential
variability between different comparisons corresponding to the columns of
the design matrix. A measure of variability is calculated for each CpG in
each sample by subtracting out the group mean and taking the absolute or
squared deviation. A linear model is then fitted to the absolute or squared
deviations. The residuals of the linear model fit are subjected to empirical
Bayes shrinkage and moderated t statistics (Smyth, 2004) calculated. False
discovery rates are calculated using the method of Benjamini and Hochberg
(1995).
Please always specify the coef
parameter in the call to varFit
,
which indicates which groups are to be tested for differential variability.
If coef
is not specified, then group means are estimated based on all
the columns of the design matrix and subtracted out before testing for
differential variability. If the design matrix contains nuisance parameters,
then subsetting the design matrix columns by coef
should remove these
columns from the design matrix. If the design matrix includes an intercept
term, this should be included in coef
. The nuisance parameters are
included in the linear model fit to the absolute or squared deviations, but
should not be considered when subtracting group means to obtain the
deviations. Note that design matrices without an intercept term are
permitted, and specific contrasts tested using the function
contrasts.varFit
.
For methylation data, the analysis is performed on the M-values, defined as
the log base 2 ratio of the methylated signal to the unmethylated signal. If
a MethylSet
object is supplied, M-values are extracted with an offset
of 100 added to the numerator and denominator.
For testing differential variability on RNA-Seq data, a DGEList
object can be supplied directly to the function. A voom
transformation is applied before testing for differential variability. The
weights calculated in voom
are used in the linear model fit.
Since the output is of class MArrayLM
, any functions that can be
applied to fit objects from lmFit
and eBayes
can be applied,
for example, topTable
and decideTests
.
Produces an object of class MArrayLM
(see
MArrayLM-class
) containing everything found in a fitted model
object produced by lmFit
and eBayes
as well as a vector
containing the sample CpG-wise variances and a matrix of LogVarRatios
corresponding to the differential variability analysis.
NULL
NULL
NULL
Belinda Phipson
Phipson, B., and Oshlack, A. (2014). A method for detecting differential variability in methylation data shows CpG islands are highly variably methylated in cancers. Genome Biology, 15:465.
Smyth, G.K. (2004). Linear models and empirical Bayes methods for assessing differential expression in microarray experiments. Statistical Applications in Genetics and Molecular Biology, Volume 3, Article 3.
Smyth, G. K. (2005). Limma: linear models for microarray data. In: Bioinformatics and Computational Biology Solutions using R and Bioconductor. R. Gentleman, V. Carey, S. Dudoit, R. Irizarry, W. Huber (eds), Springer, New York, 2005.
Benjamini, Y., and Hochberg, Y. (1995). Controlling the false discovery rate: a practical and powerful approach to multiple testing. Journal of the Royal Statistical Society Series, B, 57, 289-300.
contrasts.varFit
, topVar
,
getLeveneResiduals
, lmFit
, eBayes
,
topTable
, decideTests
, voom
# Randomly generate data for a 2 group problem with 100 CpG sites and 5
# arrays in each # group.
y<-matrix(rnorm(1000),ncol=10)
group<-factor(rep(c(1,2),each=5))
design<-model.matrix(~group)
# Fit linear model for differential variability
vfit<-varFit(y,design,coef=c(1,2))
# Look at top table of results
topVar(vfit,coef=2)
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.