Robust Dynamic Variable Selection for SNP Genotype Classification

• Activity Number: 524
• Type: Contributed
• Date/Time: Wednesday, August 4, 2010 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistical Learning and Data Mining
• Author(s): Mohua Podder*+ and William J. Welch and Ruben H. Zamar and Scott J. Tebbutt
• Companies: The University of British Columbia and The University of British Columbia and The University of British Columbia and Hogg iCAPTURE Centre for Cardiovascular and Pulmonary Research
• Keywords: ensemble of classifiers ; mixture model ; genotyping ; microarray ; single nucleotide polymorphism ; outlier

Abstract

We present methodology and algorithms for classification of single nucleotide polymorphism (SNP) genotypes. A mixture model for classification provides robustness against outlying values of the explanatory variables. Furthermore, different sets of explanatory variables are generated by deliberate redundancy in the genotyping chemistry, and the algorithm chooses among these sets in a dynamic way, prediction by prediction.

For 100 SNPs, using a set of training (SIRS) samples we tested our model on a separate set of test (Coriell) samples. The overall accuracy was 99.44% with 100% predictability.

Though motivated by the genotyping application, the proposed methodology would apply to other classification problems where the explanatory variables fall naturally into groups, or outlier in the explanatory variables require variable selection at the prediction stage for robustness.