MACLEAPS

MACLEAPS applies support vector machines (SVMs) to genome-wide association study (GWAS) single-nucleotide polymorphism (SNP) datasets to predict disease risk and evaluate contributions of common, uncommon (1–5% frequency) and rare (<1% frequency) variants.

Key Features:

• Machine Learning Integration: Uses support vector machines (SVMs) to model genetic predictors from GWAS SNP datasets.
• Predictive Modeling: Constructs and evaluates predictive models for diseases including Parkinson's disease (PD) and type 1 diabetes (T1D) using area under the receiver operating characteristic curve (AUC) as a performance metric.
• Cross-Validation Techniques: Implements traditional training-validation splits and nested k-fold cross-validation for model evaluation and generalization assessment.
• Inclusion of Rare Variants: Incorporates uncommon (1–5% frequency) and rare (<1% frequency) variants in analyses to assess their impact on prediction accuracy.
• Simulation-based Evaluation: Performs simulations to investigate how effect size magnitude and heritability influence predictive performance.

Scientific Applications:

• Type 1 diabetes (T1D): Applied to T1D datasets, achieving an AUC of approximately 0.88 and estimating heritability near 90%.
• Parkinson's disease (PD): Applied to PD datasets, yielding an AUC of approximately 0.56 and estimating heritability near 38%.

Methodology:

Computational methods explicitly include SVM-based modeling of GWAS SNP data with explicit handling of common, uncommon (1–5%) and rare (<1%) variants, simulation studies to probe effect size and heritability impacts, and model evaluation via training-validation splits and nested k-fold cross-validation using AUC as the performance metric.