AIMSmodel: The AIMS model

Description

This is the model definition for AIMS. It contains the naive bayes classifier composed of the 100 rules described in Paquet et al. "Absolute assignment of breast cancer intrinsic molecular subtype" (under review at JNCI).

Usage

AIMSmodel

Arguments

Value

all.pairsThe 100 rules in AIMS in the form EntrezID gene A < EntrezID gene B
kThe selected number of optimal rules. For AIMS we have shown it is 20.
one.vs.all.tspThe Naive bayes classifier used in combination with the 100 rules
selected.pairs.listThe list of rules sorted from the best discriminating rule to the least discriminating rules subdivided by subtype.

Details

This is the AIMS model define using 100 simple rules of the form gene A < gene B and combine within a naive bayes classifier within e1071. (Paquet et al. under review JNCI)

Briefly, using a suitably large training dataset(~5000 gene breast cancer gene expression profiles), the approach identifies a small set of simple binary rules (~20) that examine the raw expression measurements for pairs of genes from a single breast cancer patient, and only that patient. The binary rules are of the form "if the expression of gene x is greater than gene y, then tend to assign subtype z for that patient". Subtypes could be : Basal, Her2, LumA, LumB, or Normal. The collection of binary rules is combined for a single estimation of a patient subtype via a single probabilistic model using naiveBayes in e1071. In this way, since only expression levels of genes with a single patient is considered, the method represents a promising approach to ablate the instability caused by relativistic approaches (Paquet et al. in review at JNCI).

Examples

Run this code

## Load a sample of the McGill dataset used in the paper
data(AIMSmodel)

## list the top-scoring rules for the individual subtypes
AIMSmodel$selected.pairs.list

## List the posterior probability tables for the 100-rules
AIMSmodel$one.vs.all.tsp[[20]]$table

Run the code above in your browser using DataLab