Bosco: Boosting Corrections for Genome-Wide Association Studies With Imbalanced Samples

Area of Science:

• Genetics
• Bioinformatics
• Machine Learning

Background:

• Genome-wide association studies (GWAS) often face imbalanced sample sizes, with more control than case samples.
• This imbalance is particularly problematic for rare diseases or rare populations, potentially biasing results and reducing power.
• Existing GWAS methods can struggle with severe statistical biases towards the majority group, hindering the identification of true disease-associated loci.

Purpose of the Study:

• To introduce Bosco, a novel boosting correction method designed to address sample imbalance in GWAS.
• To enhance the statistical power of GWAS for detecting disease-associated genetic variants in imbalanced datasets.
• To validate Bosco's performance using simulated data and a real-world genome-scale gastric cancer dataset.

Main Methods:

• Bosco employs a coarse-to-fine learning framework inspired by machine learning boosting theory.
• The coarse step assigns importance scores to samples in the major group (controls).
• The fine step calculates P-values using weighted logistic regression, correcting for sample imbalance.

Main Results:

• Bosco significantly improves discovery power on extremely imbalanced datasets while maintaining control over false positives.
• The method successfully replicated known findings in a gastric cancer GWAS dataset with high statistical significance.
• Bosco demonstrated the potential to identify novel single nucleotide polymorphisms (SNPs) associated with gastric cancer.

Conclusions:

• Bosco effectively mitigates statistical biases caused by imbalanced sample sizes in GWAS.
• The method offers a powerful tool for uncovering true genetic associations, especially in scenarios with limited case data.
• Bosco shows promise for advancing genetic research in rare diseases and underrepresented populations.