Gradient Boosting as a SNP Filter: an Evaluation Using Simulated and Hair Morphology Data.

Area of Science:

• Genetics
• Statistical genetics
• Bioinformatics

Background:

• Genome-wide association studies (GWAS) typically use simple additive models for single nucleotide polymorphisms (SNPs).
• The computational burden of testing all possible genetic models (e.g., recessive, dominant, SNP-SNP, SNP-environment interactions) is prohibitive for genome-wide data.
• Existing GWAS methods struggle to detect complex interactions, especially when one interacting SNP has no main effect.

Purpose of the Study:

• To propose and evaluate a two-step approach for GWAS using a sensitive filtering method.
• To assess the utility of Gradient Boosting Machine (GBM) as a filter for detecting various SNP effects and interactions.
• To enable more feasible in-depth analysis of selected SNPs by reducing the overall number of variants.

Main Methods:

• Utilized Gradient Boosting Machine (GBM), a machine learning algorithm, as a primary filtering step.
• GBM was evaluated for its ability to detect SNP main effects and various interaction types without pre-specified genetic models.
• Included a large number of covariates to explore gene-environment (GxE) interactions.
• Performed simulations to compare GBM performance against standard additive regression models.
• Analyzed empirical data on hair morphology.

Main Results:

• GBM demonstrated strong performance, even in scenarios favorable to standard GWAS additive models.
• GBM effectively detected interaction effects, including those where one interacting SNP had a zero main effect, which are missed by traditional GWAS.
• The analysis of hair morphology data indicated that selecting 10K-20K top-ranked SNPs in the first step is sufficient for explaining phenotypic variance.
• GBM facilitates the exploration of multiple GxE interactions, overcoming limitations of parametric GWAS frameworks.

Conclusions:

• A two-step GWAS approach employing GBM as a filter is a powerful strategy for efficiently identifying significant SNPs and complex genetic interactions.
• GBM offers a computationally feasible method for exploring diverse genetic effects and interactions in large-scale genomic datasets.
• This approach enhances the power of GWAS to detect biologically relevant genetic architectures, including complex GxE interactions.