Combining associative classification with multifactor dimensionality reduction for predicting higher-order SNP interactions in case-control studies

Abstract

The identification and characterisation of genotype-phenotype mapping is a central focus of current genome wide association interaction studies (GWAIS). Revealing these relationships for exposing the hidden structures of diseases has received considerable attention by a number of researchers. However, the current statistical and computational approaches ignore many complex genetic contexts. A multifactor dimensionality reduction based on associative classification was previously proposed for detecting multi-locus single nucleotide polymorphism (SNP) interactions in GWAIS. The approach is further studied in detail by adjusting threshold levels, and adding noise to the datasets. The simulated studies demonstrated significant improvements in accuracy by adjusting threshold values over the previous approaches. The results also indicate that the approach is robust in the presence of noise. Further, the application of this approach to real world data has demonstrated higher-order interactions among five SNPs for the manifestation of breast cancer, and three SNPs for the manifestation of hypertension.