semisup

semisup implements a parametric semi-supervised mixture model to detect main and interaction effects of single nucleotide polymorphisms (SNPs) on quantitative traits by testing each SNP for evidence of participation in interactions, including with unobserved factors.

Key Features:

• Parametric semi-supervised mixture model: Uses a parametric semi-supervised mixture formulation to combine signals of main and interaction effects on quantitative traits.
• Single-SNP testing: Tests each SNP individually for participation in any interaction, reducing the multiple-testing burden to one test per SNP.
• Genotype partitioning: Partitions individuals by genotype defined as the number of minor alleles.
• Dual-effect evaluation: Evaluates (i) difference in group means as a main effect and (ii) difference in the distribution of trait values among a subset of individuals as an interaction signal.
• Semi-supervised mixture test integration: Integrates the main-effect and interaction-effect components into a single semi-supervised mixture test.
• Posterior membership probabilities: Produces posterior membership probabilities that can be used to indirectly suggest interacting variables.
• Statistical performance: Demonstrates high statistical power and valid control of type I error on simulated and real datasets.
• Applicability: Applicable to genome-wide association studies (GWAS) and differential expression settings without requiring explicit enumeration of interaction terms.

Scientific Applications:

• GWAS marker detection: Detects markers with main or interactive effects in genome-wide association studies.
• Differential expression analysis: Identifies interaction-associated signals in differential expression settings.
• Detection of interactions with unobserved factors: Tests for SNPs that interact with unobserved or unmeasured variables by assessing distributional deviations.
• Method validation: Applied to simulated and real datasets to evaluate statistical power and type I error control.

Methodology:

Partition individuals by genotype (number of minor alleles), evaluate (i) differences in group means and (ii) differences in the distribution of trait values among a subset of individuals, integrate these components using a semi-supervised mixture test, and compute posterior membership probabilities.