An ensemble learning approach jointly modeling main and interaction effects in genetic association studies

Area of Science:

• Genetics
• Computational Biology
• Statistical Genomics

Background:

• Complex diseases arise from intricate interactions between multiple genes and environmental factors.
• Identifying these genetic contributors requires methods capable of analyzing gene-gene interactions across the genome.
• Current analytical approaches often struggle to capture the joint effects of multiple genetic loci.

Purpose of the Study:

• To introduce a novel ensemble learning approach (ELA) for detecting sets of genetic loci with significant joint main and interaction effects on complex traits.
• To develop a method that provides interpretable results, including a final model, overall association P-value, and importance measures for loci and their interactions.
• To evaluate the performance of ELA against existing methods using simulations and real-world genetic data.

Main Methods:

• The ensemble learning approach (ELA) involves identifying "base learners" representing individual gene effects or interactions.
• These base learners are subsequently combined using a linear model to form a final predictive model.
• The approach generates an overall P-value for the set of identified loci and measures the importance of each locus and interaction.

Main Results:

• Simulation studies confirmed ELA's superior power compared to single-marker tests across all scenarios.
• ELA outperformed three other multi-locus methods in most tested cases.
• In a Type 2 diabetes study, ELA identified 11 single nucleotide polymorphisms with significant joint effects, which were missed by marginal and two-locus interaction tests.

Conclusions:

• The proposed ensemble learning approach (ELA) is a powerful and interpretable tool for analyzing complex diseases.
• ELA effectively identifies multi-locus genetic effects, offering advantages over traditional single-marker and pairwise interaction analyses.
• ELA demonstrates significant potential for discovering novel genetic associations in large-scale genetic studies, as evidenced by its application to Type 2 diabetes.