smoothExp

0th

Percentile

smooth gene expression trajectories in an ExpressionSet, using a phenoData variable as the domain

smooth gene expression trajectories in an ExpressionSet, using a phenoData variable as the domain

Keywords
models
Usage
smoothExp(eset, phenovar, keepAt = NULL, fitter = mgcv::gam,
                 fmla = y ~ s(x, k = 9, fx = TRUE), retain =
                 c("gcv.ubre", "sig2", "coefficients"), igran = 50,
                 exclind = NULL, dogcv.locfit = FALSE, maxna = 5, pv =
                 varfitRanpart(fitter, fmla))
smoothExpFull(eset, phenovar, fitter, fmla=y ~ s(x, k = 9, fx=TRUE))
Arguments
eset
instance of ExpressionSet
phenovar
string naming a phenoData variable in eset that will serve as the domain of the expression trajectory function
keepAt
sequence of values of domain variable at which fitted values are to be computed and saved
fitter
R function that will fit a model given a formula as supplied in fmla
fmla
A formula object that will serve as a template; variable y is bound to expression values, and x is bound to domain values
retain
names of extra components of the fitted model object to be extracted and saved
igran
granularity of the iteration count report (string emitted if options()$verbose is TRUE
dogcv.locfit
set to TRUE to obtain the locfit gcv estimate
exclind
if pv is varpredFIX a fixed set of test point indices is specified in this vector; leave NULL otherwise
maxna
numeric threshold, maximum number of NA values of expression permitted; if there are more, the gene is skipped
pv
predictive variance plugin -- accepts a closure
Details

This function iterates over all features in an ExpressionSet instance, fitting smooth models for expression as a function of some phenodata variable, typically a time measure. The objective is to support a ``drop-in'' interface, so that alternate smoothing technologies may be incorporated.

Value

  • a list with one element per gene with components origx, origy, newx, preds and extras for the retain components, and locfit.gcv if dogcv.locfit is set.

Note

A related function smoothExpFull is available that retains the full fitted model object -- this can be very unwieldy but has good downstream functionality; arbitrary predictions can be obtained, for example.

Aliases
  • smoothExpFull
  • smoothExp
Examples
data(yccalp)
okp = options()
options(verbose=TRUE)
dem = smoothExp( yccalp[1:20,], "minutes", keepAt=seq(0,120,5), igran=5)
plot(dem[[1]])
dem2 = smoothExp( yccalp[1:20,], "minutes", keepAt=seq(0,120,5),
   fitter=locfit::locfit, fmla=y~lp(x,nn=.4),dogcv.locfit=TRUE, igran=5)
plot(dem2[[1]])
dem3 = smoothExp( yccalp[1:20,], "minutes", keepAt=seq(0,120,5),
   fitter=locfit::locfit, fmla=y~lp(x,nn=.4),dogcv.locfit=FALSE,
   exclind=c(3,6,9,12,15), pv=varfitFIX(locfit::locfit,
   y~lp(x,nn=.4), c(3,6,9,12,15)), igran=5)
plot(dem3[[1]])
pv2 = gcvs(dem2) # get locfit gcv est
pv3 = extract(dem3, "predvar") # use the fixed test points est
ipar = par(no.readonly=TRUE)
par(mfrow=c(2,2))
plot(pv2,pv3)
pv3
bes3 = names(sort(pv3))[1]
plot(dem2[[bes3]], main="full data fit")
plot(dem3[[bes3]], main="leave 5 out fit, x=test point")
par(ipar)

options(okp)
Documentation reproduced from package smoothMiner, version 0.1-10, License: Artistic-2.0

Community examples

Looks like there are no examples yet.