An enhanced machine learning tool for cis-eQTL mapping with regularization and confounder adjustments.

Area of Science:

• Genetics
• Bioinformatics
• Machine Learning

Background:

• Expression quantitative trait loci (eQTL) studies investigate gene regulation variants.
• Current eQTL methods often lack statistical power and struggle with multiple weak genetic variants.
• Existing algorithms may not effectively model genetic and non-genetic confounding factors.

Purpose of the Study:

• To introduce a novel machine learning algorithm, LSKM-LASSO, for robust eQTL identification.
• To address limitations in modeling joint effects of multiple genetic variants and confounding factors.
• To improve the accuracy of eQTL mapping and phenotype prediction.

Main Methods:

• Developed LSKM-LASSO, a machine learning algorithm combining least-squares kernel machine with a LASSO regularization term.
• Modeled the simultaneous association between multiple genetic variants and phenotypic traits.
• Incorporated a flexible framework for estimating genetic confounding.

Main Results:

• LSKM-LASSO demonstrated superior performance over state-of-the-art methods in simulations for eQTL identification and phenotype prediction.
• Applied to Genotype-Tissue Expression (GTEx) data across 11 tissues.
• Identified a greater number of genes with significant eQTLs compared to existing algorithms.

Conclusions:

• LSKM-LASSO offers a powerful and flexible tool for eQTL mapping.
• The method enhances the understanding of genetic architecture underlying complex traits.
• Improved eQTL identification facilitates more accurate phenotype prediction.