A model-based approach to selection of tag SNPs

Area of Science:

• Genomics
• Bioinformatics
• Statistical Genetics

Background:

• Single Nucleotide Polymorphisms (SNPs) are common variations in the human genome.
• Identifying tag SNPs is crucial for efficient genetic association studies by capturing haplotype information.
• Optimal tag SNP selection requires robust probabilistic models that accurately represent Linkage Disequilibrium (LD).

Purpose of the Study:

• To develop and evaluate tag SNP selection methods based on entropy maximization and probabilistic models.
• To compare the performance of different models, including the Li and Stephens hidden Markov model, for tag SNP selection.
• To assess the informativeness and predictive ability of selected tag SNP sets.

Main Methods:

• Computed description code-lengths of SNP data using various probabilistic models.
• Developed tag SNP selection strategies based on entropy maximization.
• Utilized datasets from the HapMap and ENCODE projects for model evaluation.

Main Results:

• The Li and Stephens hidden Markov model demonstrated superior performance in describing SNP data and maximizing information content.
• Tag SNP sets selected using this model showed enhanced ability to predict tagged SNPs compared to other methods.
• Information content, evaluated with a good model, proved more sensitive for assessing tag set quality than prediction rates.

Conclusions:

• Tag sets selected via the Li and Stephens model significantly outperform those from existing methods.
• Haplotype informativeness, assessed through robust models, is key for effective tag SNP selection, even without direct haplotype genotyping.
• Haplotype phase uncertainty has minimal impact on the predictive power of well-selected tag SNP sets.