epoc (version 0.2.6-1.1)

EPoC: EPoC

Description

EPoC (Endogenous Perturbation analysis of Cancer)

Usage

epocA(Y, U=NULL, lambdas=NULL, thr=1.0e-10, trace=0, ...)
epocG(Y, U, lambdas=NULL, thr=1.0e-10, trace=0, ...)
epoc.lambdamax(X, Y, getall=F, predictorix=NULL)
as.graph.EPOCA(model,k=1)
as.graph.EPOCG(model,k=1)
write.sif(model, k=1, file="", append=F)
# S3 method for EPOCA
print(x,...)
# S3 method for EPOCG
print(x,...)
# S3 method for EPOCA
summary(object, k=NULL, ...)
# S3 method for EPOCG
summary(object, k=NULL,...)
# S3 method for EPOCA
coef(object, k=1, ...)
# S3 method for EPOCG
coef(object, k=1, ...)
# S3 method for EPOCA
predict(object, newdata,k=1,trace=0, ...)
# S3 method for EPOCG
predict(object, newdata,k=1,trace=0, ...)

Arguments

Y

N x p matrix of mRNA transcript levels for p genes and N samples for epocA and epocG. For epoc.lambdamax Y is a multi-response matrix

U

N x p matrix of DNA copy number

lambdas

Non-negative vector of relative regularization parameters for lasso. \(\lambda=0\) means no regularization which give dense solutions (and takes longer to compute). Default=NULL means let EPoC create a vector

thr

Threshold for convergence. Default value is 1e-10. Iterations stop when max absolute parameter change is less than thr

trace

Level of detail for printing out information as iterations proceed. Default 0 -- no information

X

In epoc.lambdamax X is the design matrix, i.e. predictors

predictorix

For epoc.lambdamax when using a multi-response matrix Y predictors are set to zero for each corresponding response. predictorix tells which of the responses that have a corresponding predictor in the network case

getall

Logical. For epoc.lambdamax get a vector of all inf-norms instead of a single maximum

file

either a character string naming a file or a connection open for writing. "" indicates output to the console

append

logical. Only relevant if file is a character string. If TRUE, the output is appended to the file. If FALSE, any existing file of the name is destroyed

model

Model set from epocA or epocG

k

Select a model of sparsity level k in [1,K]. In summary default (NULL) means all. In plot default is first model.

newdata

List of Y and U matrices required for prediction. epocG requires just U.

x

Model parameter to print and plot

object

Model parameter to summary, coef and predict

...

Parameters passed down to underlying function, e.g. print.default. For epocA and epocG … are reserved for experimental options.

Value

epocA and epocG returns an object of class `"epocA"' and `"epocG"' respectively.

The methods summary, print, coef, predict can be used as with other models. coef and predict take an extra optional integer parameter k (default 1) which gives the model at the given density level.

An object of these classes is a list containing at least the following components:

coefficients

list of t(A) or t(G) matrices for the different \(\lambda\)s

links

the number of links for the different \(\lambda\)s

lambdas

the \(\lambda\)s used for this model

%\item{fitted.values}{the fitted mean values} %\item{residuals}{the residuals, i.e. the response minus fitted values}

R2

R<U+00B2>, coefficient of determination

Cp

Mallows Cp

%\item{R2.adj}{adjusted R<U+00B2>, coefficient of determination adjusted for number of explanatory terms}

s2

Estimate of the error variance

RSS

Residual Sum of Squares (SSreg)

SS.tot

Total sum of squares of the response

inorms

the infinity norm of predictors transposed times response for the different responses

d

Direct effects of CNA to corresponding gene

Details

epocA and epocG estimates sparse matrices \(A\) or \(G\) using fast lasso regression from mRNA transcript levels \(Y\) and CNA profiles \(U\). Two models are provided, EPoC A where

$$AY + U + R = 0$$

and EPoC G where

$$Y = GU + E.$$

The matrices \(R\) and \(E\) are so far treated as noise. For details see the reference section and the manual page of lassoshooting.

If you have different sizes of U and Y you need to sort your Y such that the U-columns correspond to the first Y-columns. Example: Y.new <- cbind(Y[,haveCNA], Y[, -haveCNA]) CHANGES: predictorix used to be a parameter with a vector of a subset of the variables 1:p of U corresponding to transcripts in Y, Default was to use all which mean that Y and U must have same size.

epoc.lambdamax returns the maximal \(\lambda\) value in a series of lasso regression models such that all coefficients are zero.

plot if type='graph' (default) plot graph of model using the Rgraphviz package arrows only tell direction, not inhibit or stimulate. If type='modelsel' see modelselPlot.

References

Rebecka J<U+00F6>rnsten, Tobias Abenius, Teresia Kling, Linn<U+00E9>a Schmidt, Erik Johansson, Torbj<U+00F6>rn Nordling, Bodil Nordlander, Chris Sander, Peter Gennemark, Keiko Funa, Bj<U+00F6>rn Nilsson, Linda Lindahl, Sven Nelander. (2011) Network modeling of the transcriptional effects of copy number aberrations in glioblastoma. Molecular Systems Biology 7 (to appear)

See Also

print, modelselPlot, epoc.validation, epoc.bootstrap, plot.EPoC.validation.pred, plot.EPoC.validation.W, coef, predict

Examples

# NOT RUN {
modelA <- epocA(X,U)
modelG <- epocG(X,U)

# plot sparsest A and G models using the igraph package
# arrows only tell direction, not inhibit or stimulate
par(mfrow=c(1,2))
plot(modelA)
plot(modelG)

# OpenGL 3D plot on sphere using the igraph and rgl packages
plot(modelA,threed=T)

# Write the graph to a file in SIF format for import in e.g. Cytoscape
write.sif(modelA,file="modelA.sif")

# plot graph in Cytoscape using Cytoscape XMLRPC plugin and 
# R packages RCytoscape, bioconductor graph, XMLRPC
require('graph')
require('RCytoscape')
g <- as.graph.EPOCA(modelA,k=5)
cw <- CytoscapeWindow("EPoC", graph = g)
displayGraph(cw)

# prediction
N <- dim(X)[1]
ii <- sample(1:N, N/3)

modelG <- epocG(X[ii,], U[ii,])
K <- length(modelA$lambda) # densest model index index
newdata <- list(U=U[-ii,])
e <- X[-ii,] - predict(modelA, newdata, k=K)
RSS <- sum(e^2)
cat("RMSD:", sqrt(RSS/N), "\n")

# }