analysisPipeline: Training of thresholds

In sigsquared: Gene signature generation for functionally validated signaling pathways

Description

The analysisPipeline function is used to train a set of thresholds for predicting survival outcome within the context of a given signaling environment. This signaling environment is encoded in a geneSignature object.

Usage

analysisPipeline(dataSet, geneSig, iterPerK=2500, k=3, rand=TRUE, newjpdf=FALSE, jpdf=FALSE, nJPDF=12500, disc=c(0.005, 0.01, 0.03, 0.05), MFS="MFS", met="met", optMeth="Nelder-Mead")

Arguments

dataSet	ExpressionSet object containing both expression data (exprs) and phenotypic survival data (pData)
geneSig	geneSignature object containing directions, thresholds, and gene symbols
iterPerK	integer number of optimization iterations for each k
k	integer k for k-fold cross-validation
rand	boolean determining whether the k subsets are randomly drawn (otherwise k subsets are selected ordinally)
newjpdf	boolean for generating a joint probability function for alternate smoothed cost function (not recommended)
jpdf	solnSpace object containing empirical joint probability function for alternate smoothed cost function (not recommended)
nJPDF	value determining the number of samples with which to estimate the empirical joint probability function for alternate smoothed cost function (not recommended)
disc	vector of discretation thresholds for discretized cost function
MFS	variable name for survival-time data in dataSet object
met	variable name for metastasis event data in dataSet object
optMeth	optimization method used by R function 'optim'

Details

The analysisPipeline function optimizes over a cost function designed to minize both type I and II error. There is a discretized and smoothed cost function available, however implementation of the smoothed cost function relies on sampling of the solution space. This sampling may be pre-computed and implemented through the 'jpdf' argument, however overall usage of the smoothed cost function is not recommended.

Value

A geneSignature object containing newly trained thresholds

Author(s)

UnJin Lee

Examples

## Load in example data
data("BrCa443")

## Create initial geneSignature object
## Note it is not necessary to define thresholds at this point
gs <- setGeneSignature(g=new("geneSignature"), direct=c(-1,1,1,1,1,1,1), genes=c("RKIP", "HMGA2", "SPP1", "CXCR4", "MMP1", "MetaLET7", "MetaBACH1"))

## Generate thresholds
gs <- analysisPipeline(dataSet=BrCa443, geneSig=gs, iterPerK=50, k=2, rand=FALSE)

sigsquared documentation built on Nov. 8, 2020, 6:59 p.m.