Exploiting SNP correlations within random forest for genome-wide association studies.

Area of Science:

• Genetics
• Bioinformatics
• Computational Biology

Background:

• Genome-wide association studies (GWAS) aim to identify genetic variants associated with traits or diseases.
• Standard GWAS methods often use univariate tests, failing to capture complex genetic interactions and linkage disequilibrium.
• Discovering multivariate genetic effects is crucial for a comprehensive understanding of disease etiology.

Purpose of the Study:

• To propose an extension of the Random Forest algorithm, named T-Trees, specifically designed for structured GWAS data.
• To evaluate the performance of T-Trees in predicting disease risk and identifying relevant genetic loci.
• To investigate the impact of quality control on predictive power and locus identification in GWAS.

Main Methods:

• Developed T-Trees, a novel Random Forest algorithm extension for structured GWAS data.
• Empirically tested T-Trees on multiple GWAS datasets.
• Assessed variable importance measures derived from T-Trees for locus identification.

Main Results:

• T-Trees significantly outperformed standard Random Forest and linear models in risk prediction on GWAS datasets.
• Empirical results suggest the existence of multivariate non-linear genetic effects from combinations of single nucleotide polymorphisms (SNPs).
• Variable importance from T-Trees effectively aids in identifying relevant genetic loci and highlights the importance of quality control.

Conclusions:

• The T-Trees method offers a powerful approach for discovering complex genetic interactions in GWAS.
• Multivariate non-linear effects play a significant role in disease risk, which can be captured by advanced algorithms like T-Trees.
• Rigorous quality control is essential for maximizing the predictive power and accuracy of genetic locus identification in GWAS.