FindBestTransform: Find the power transformation that makes a data set...

In PoiClaClu: Classification and Clustering of Sequencing Data Based on a Poisson Model

Description

Find a constant alpha, 0<alpha<=1, such that x raised to the power alpha approximately follows the simple Poisson log linear model that says that the (i,j) element of x is Poisson with mean si times gj, where si is a sample-specific term and gj is a feature-specific term. Alpha is selected via a grid search.

Usage

FindBestTransform(x)

Arguments

x	A n-by-p matrix of sequencing data, with n observations and p features.

Value

Returns alpha, the power to which x should be raised.

Author(s)

Daniela Witten

References

D Witten (2011) Classification and clustering of sequencing data using a Poisson model. To appear in Annals of Applied Statistics.

Examples

set.seed(1)
dat <- CountDataSet(n=20,p=100,sdsignal=2,K=4,param=10)
alpha <- FindBestTransform(dat$x)
# This is the best transformation!
dd <- PoissonDistance(dat$x^alpha,type="mle", transform=FALSE)
# OR we could get the samething automatically:
dd2 <- PoissonDistance(dat$x,type="mle",transform=TRUE)
# or like this:
dd3 <- PoissonDistance(dat$x,type="mle",transform=TRUE,alpha=alpha)
ColorDendrogram(hclust(dd$dd), y=dat$y, branchlength=10)

PoiClaClu documentation built on May 2, 2019, 8:29 a.m.