Statistical Learning Methods Applicable to Genome-Wide Association Studies on Unbalanced Case-Control Disease Data.

Area of Science:

• Genetics
• Bioinformatics
• Statistical Genomics

Background:

• Imbalanced case-control ratios are common in genome-wide association studies (GWAS), particularly for low-prevalence diseases.
• The increasing size of biobanks and electronic health records exacerbates this issue.
• Unbalanced binary traits challenge traditional statistical methods like linear mixed models (LMM), leading to inflated type I error rates.

Purpose of the Study:

• To review existing statistical approaches for handling unbalanced case-control data in GWAS.
• To evaluate the advantages and limitations of these methods.
• To explore the potential application of novel machine learning techniques in GWAS with imbalanced data.

Main Methods:

• Literature review of statistical methods for unbalanced GWAS data.
• Analysis of the performance and limitations of established approaches.
• Exploration of state-of-the-art machine learning algorithms for potential GWAS application.

Main Results:

• Traditional methods like LMM show limitations with unbalanced data, yielding inflated type I errors.
• Various statistical strategies exist to mitigate inaccuracies caused by case-control imbalance.
• Machine learning approaches offer promising, yet unexplored, avenues for analyzing imbalanced GWAS data.

Conclusions:

• Addressing case-control imbalance is crucial for accurate GWAS.
• Existing statistical methods have trade-offs that require careful consideration.
• Further research into machine learning applications could significantly advance GWAS of imbalanced traits.