MR-MDR

MR-MDR identifies gene-gene interactions that contribute to multivariate continuous phenotypes in genome-wide association studies by applying rank-based nonparametric methods that are robust to skewed distributions and outliers.

Key Features:

• Robustness to distributional assumptions: Employs nonparametric spatial signs and ranks instead of Hotelling's T² to reduce sensitivity to skewed data and outliers.
• Handling multiple continuous phenotypes: Supports analysis of correlated multivariate continuous phenotypes to detect interaction effects across traits.
• Fuzzy k-means clustering: Uses fuzzy k-means to classify multi-locus genotypes into two groups for downstream evaluation.
• Spatial rank-sum statistic: Uses a spatial rank-sum statistic as the evaluation measure and selects the best interaction model based on the largest statistic.
• Cross-validation: Incorporates tenfold cross-validation to mitigate overfitting during model selection.

Scientific Applications:

• GWAS of kidney-related phenotypes: Applied to a Korean genome-wide association study to identify genetic interactions associated with four phenotypes related to kidney function.
• Simulation-based performance evaluation: Simulation studies demonstrated superior performance for skewed phenotype distributions and comparable power to other methods for symmetric distributions across varying phenotype correlations and sample sizes.

Methodology:

Cluster multi-locus genotypes using fuzzy k-means, compute spatial signs and ranks and the spatial rank-sum statistic to evaluate models, select the model with the largest statistic, and assess generalizability with tenfold cross-validation.