Identification of interactions using model-based multifactor dimensionality reduction.

Area of Science:

• Genetics
• Bioinformatics
• Machine Learning

Background:

• Complex traits arise from multiple genetic and environmental factors.
• Machine learning methods are valuable for identifying gene-gene interactions without strict genetic model assumptions.
• Filter methods aim to reduce computational burden by pre-selecting relevant factors.

Purpose of the Study:

• To apply model-based multifactor dimensionality reduction (MB-MDR) to Genetic Analysis Workshop 19 (GAW19) data.
• To identify interaction effects on various traits, including hypertension status, systolic blood pressure (SBP), and diastolic blood pressure (DBP).
• To assess the compatibility of nearest neighbor-based filtering methods with MB-MDR.

Main Methods:

• Model-based multifactor dimensionality reduction (MB-MDR), a semiparametric machine learning approach, was utilized.
• MB-MDR allows for adjustment of confounding variables and lower-level effects.
• Nearest neighbor-based filtering methods were evaluated for their compatibility with MB-MDR.

Main Results:

• A significant interaction between SNP rs859400 and age was observed for SBP.
• Identified 23 SNP-SNP interactions for hypertension status, 42 for SBP, and 26 for DBP.
• Some identified SNPs were in strong linkage disequilibrium (LD); three interactions for hypertension status were found in filtered subsets.

Conclusions:

• The evaluated filtering methods demonstrated incompatibility with MB-MDR.
• Linkage disequilibrium (LD) pruning is recommended as a quality control step to manage computational complexity by removing redundant SNPs.