A Hybrid feature selection model for genome wide association studies

Download
2013
Yücebaş, Sait Can
Through Genome Wide Association Studies (GWAS) many SNP-complex disease relations have been investigated so far. GWAS presents high amount – high dimensional data and relations between SNPs, phenotypes and diseases are most likely to be nonlinear. In order to handle high volume-high dimensional data and to be able to find the nonlinear relations, data mining approaches are needed. A hybrid feature selection model of support vector machine and decision tree has been designed. This model also combines the genotype and phenotype information to increase the diagnostic performance. The model is tested on prostate cancer and melanoma data that have been downloaded from NCBI’s dbGaP database. On prostate cancer data the hybrid system performed 71.67% accuracy on data set consists of only genotypes, 84.23% accuracy on data set consists of only phenotypes and when genotyping and phenotypes are integrated accuracy increased to 93.81%. On melanoma data, the hybrid system performed 57.12% accuracy for only genotypes, 75.48% accuracy for only phenotypes and when genotyping and phenotypes are integrated accuracy increased to 86.35%. For prostate cancer case the hybrid system’s has performance indicators of 90.92% of sensitivity and 0.91 AUC, which outperforms Prostate Specific Antigen (PSA) test. In melanoma case selected phenotypic and genotypic features were also examined by previous studies that shows the ability of the system to select most predictive features so the hybrid system on melanoma case has a potential to be used for identifying the risk groups.
Citation Formats
S. C. Yücebaş, “A Hybrid feature selection model for genome wide association studies,” Ph.D. - Doctoral Program, Middle East Technical University, 2013.