Detecting SNP–SNP Interactions in Imbalanced Case-Control Study

SNP–SNP interactions are particularly informative biomarkers regarding the genetic components of disease risk. However, SNP–SNP interaction identifications are yet limited in imbalanced case–control study. In this study, we proposed a multiobjective multifactor dimensionality reduction (MOMDR) based on three balancing approaches (BMOMDR), including (1) stratified
$K$
-fold cross-validation; (2) balanced estimation of ratio between cases and controls; (3) balanced measures of SNP–SNP interactions, to effectively identify SNP–SNP interaction in imbalanced case–control study. BMOMDR was evaluated by extensive experiments on both simulated imbalanced case–control datasets and real genome-wide data from Wellcome Trust Case Control Consortium (WTCCC). For the simulated datasets, the results indicated that three balancing approaches can enhance the detection success rate of SNP–SNP interaction by MOMDR in imbalanced datasets. For WTCCC datasets, the results of SNP–SNP interaction detection obtained from BMOMDR revealed statistically significant (
$p < 0.0001$
), revealing that BMOMDR can effectively identify SNP–SNP interaction in imbalanced case–control study. BMOMDR is freely available at
http://shorturl.at/bluJS
.