NNPERM

NNPERM analyzes case-control multi-locus genotype data using an artificial neural network combined with permutation testing to detect associations between genetic markers and disease phenotypes without requiring haplotype phase information or evolutionary assumptions.

Key Features:

• Artificial Neural Network Integration: Uses an ANN to learn patterns in multi-marker genotypes and classify cases versus controls based on marker genotypes.
• Permutation Testing for Statistical Significance: Assesses significance by permuting genotype data relative to affection status and comparing the real dataset's predictive performance to numerous randomized datasets.
• Optimization of Network Parameters: Performs systematic exploration of network parameters to identify optimal ANN configurations.
• Phase-free Multi-locus Analysis: Analyzes multi-locus genotype data from biallelic markers such as SNPs without requiring haplotype phase information or evolutionary history assumptions.
• Conditional Analyses with Covariates: Enables incorporation of known risk factors alongside marker genotypes within the permutation framework to assess independent marker contributions.
• Empirical Application: Has been applied to SNPs in the calpain 10 (CAPN10) gene in case-control samples of subjects with type 2 diabetes and impaired glucose tolerance, showing improved significance versus single-marker tests adjusted for multiple testing.

Scientific Applications:

• Disease Locus Identification: Localizes disease-associated loci using multi-locus haplotypes from biallelic markers such as SNPs more effectively than single-marker analyses.
• Complex Trait Analysis: Detects associations in traits where major gene effects or multi-marker interactions may underlie disease risk.
• Integration with Known Risk Factors: Assesses genetic associations conditional on additional risk factors to separate independent genetic effects from known covariates.

Methodology:

Input multi-marker genotype data from case-control studies; train an ANN to approximate affection status; perform permutation testing by permuting genotypes relative to affection status to generate randomized datasets; compare predictive power of the real dataset to randomized datasets; systematically explore network parameters to optimize performance.