Improving genetic variant identification for quantitative traits using ensemble learning-based approaches.

Area of Science:

• Genetics
• Bioinformatics
• Computational Biology

Background:

• Genome-wide association studies (GWAS) are advancing with new genome assemblies.
• Current GWAS methods primarily focus on discrete phenotypes, leaving quantitative trait (QT) analysis underdeveloped.
• Existing methods can overlook significant variants due to multicollinearity and strict p-value thresholds.

Purpose of the Study:

• To develop and validate an enhanced ensemble learning approach for QT analysis in GWAS.
• To improve the identification of genetic variants associated with quantitative traits.
• To address limitations in current GWAS methods for QTs.

Main Methods:

• Proposed an ensemble learning approach integrating regularized variant selection and machine learning association methods.
• Benchmarked four variant selection methods (LASSO, ridge, elastic-net, mutual information) and four association methods (linear regression, random forest, SVR, XGBoost).
• Evaluated the approach on simulated and real-world datasets (PennCATH) for low-density lipoprotein (LDL)-cholesterol levels.

Main Results:

• The combination of elastic-net with Support Vector Regression (SVR) demonstrated superior performance across all tested datasets.
• Functional annotation of top identified single nucleotide polymorphisms (SNPs) revealed expression in relevant tissues for LDL cholesterol regulation.
• Confirmed involvement of known cholesterol-related genes and identified potential novel drug targets.

Conclusions:

• The developed ensemble learning approach effectively identifies genetic variants associated with quantitative traits.
• Future improvements are expected with the integration of Telomere-to-Telomere (T2T) and pangenome references in GWAS.