A comparative study on the unified model based multifactor dimensionality reduction methods for identifying gene-gene...

Area of Science:

• Genetics
• Bioinformatics
• Statistical Genetics

Background:

• Gene-gene interaction analysis is crucial for understanding complex diseases.
• Existing methods like MDR, Cox-MDR, and KM-MDR have limitations, including the need for cross-validation and permutation testing.
• The Unified Model-based Multifactor Dimensionality Reduction (UM-MDR) method simplifies interaction detection by integrating significance testing within a regression framework.

Purpose of the Study:

• To propose and evaluate two novel methods, KM-UMMDR and Cox2-UMMDR, for gene-gene interaction analysis.
• To compare the performance of KM-UMMDR, Cox2-UMMDR, and Cox-UMMDR against existing methods (Cox-MDR, KM-MDR).
• To assess the robustness of these methods in the presence of covariate effects and single nucleotide polymorphism (SNP) main effects on survival phenotypes.

Main Methods:

• Developed KM-UMMDR by combining Kaplan-Meier (KM)-MDR with UM-MDR.
• Developed Cox2-UMMDR by modifying Cox-MDR with UM-MDR, adjusting for covariates in the first step.
• Conducted simulation studies to compare the statistical power of the proposed methods and existing ones under various scenarios, including covariate effects, SNP main effects, censoring fractions, and minor allele frequencies (MAF).

Main Results:

• Cox-UMMDR demonstrated superior performance across all simulated scenarios, showing robustness to covariate and SNP main effects.
• Methods incorporating UM-MDR (Cox-UMMDR, KM-UMMDR, Cox2-UMMDR) outperformed traditional Cox-MDR and KM-MDR when SNPs had marginal effects.
• Cox-UMMDR and Cox-MDR were more effective than KM-UMMDR and KM-MDR in the presence of covariate effects, highlighting the advantage of adjusting for covariates.
• Cox2-UMMDR's performance was similar to KM-UMMDR/KM-MDR when covariate effects were present, suggesting that adjusting covariates in the second step (as in Cox-UMMDR) is more efficient.

Conclusions:

• KM-UMMDR and Cox2-UMMDR offer simplified implementation for detecting gene-gene interactions associated with survival time.
• The proposed UM-MDR-based methods (KM-UMMDR, Cox2-UMMDR, Cox-UMMDR) effectively identify significant interactions, outperforming Cox-MDR and KM-MDR, especially when marginal SNP effects could mask epistasis.
• Cox-UMMDR is identified as the most robust method for analyzing survival data with potential confounding covariate effects and SNP main effects.