New Evaluation Measures for Multifactor Dimensionality Reduction in SNP-SNP Interaction Analysis

Studies have proven that single nucleotide polymorphism (SNP)-SNP interaction detection is helpful for understanding the susceptibility of an individual to genetic diseases. Although multifactor dimensionality reduction (MDR) is an effective SNP-SNP interaction detection algorithm, the mechanism of SNP-SNP interaction detection based on MDR contingency tables has not been widely studied. In this study, we propose a multi-objective MDR to detect SNP-SNP interactions. In the proposed multi-objective MDR, multiple measures can be simultaneously considered for detecting epistatic interactions. Then, set theory is used to select the best epistatic interactions in k-fold cross-validation to achieve high identification accuracy for SNP-SNP interactions. Two MDR parameters, namely the correct classification rate (CCR) and predictive summary index (PSI), were used for evaluating the algorithms. The results revealed that the detection success rates of multi-objective MDR were higher than those of other MDR-based algorithms in identifying epistatic interactions. Based on the CCR and PSI, our study demonstrated that the proposed multi-objective MDR can effectively detect SNP-SNP interactions.