Statistics and Data Analysis for Microarrays Using R and Bioconductor, 2nd Edition

Bioconductor: principles and illustrations 203

7.3.3 Predictive modeling using microarray data

To conclude this overview of Bioconductor facilities for analysis of g enome-

scale data, we undertake a simple machine learning exer cise. We will use

Breiman’s “random forest” procedure to create a nd evaluate a classifier, a

mapping from the hu6800-based quantification of the transcriptome to the

clinical diagnosis labe ls “ALL,” “AML”. This is easily carr ie d out using the

MLInterfaces package. We will filter the expression data shar ply, to the an-

notated genes with variance across samples in the top 10% of the distribution

of variances.

> library(MLInterfaces)

> GMF = nsFilter(Golub_Merge, var.cutoff=.9)[[1]]

> rf1 = MLearn(ALL.AML~., data=GMF,

+ rando