Bayesian Gaussian Mixture Models for High-Density Genotyping Arrays.

Area of Science:

• Genetics
• Bioinformatics
• Computational Biology

Background:

• Affymetrix SNP genotyping chips offer cost-effective gene mapping but present genotype calling challenges.
• Traditional clustering methods struggle with accuracy on large, diverse datasets or when family data is unavailable.

Purpose of the Study:

• To enhance genotype calling accuracy on Affymetrix SNP data.
• To develop a robust model that leverages family relationships and empirical data for improved parameter estimation.

Main Methods:

• Constructed Gaussian mixture models with empirically derived priors for genotype calling.
• Incorporated pedigree analysis to model correlations between related individuals' signals.
• Examined symmetry hypotheses to refine parameterization and model assumptions.
• Utilized the Bayesian information criterion for model selection.

Main Results:

• The developed model demonstrated improved genotype calling compared to Affymetrix's standard software.
• Achieved a reduction in "no-calls" (uncalled genotypes) with minimal impact on overall accuracy.
• Successfully captured signal correlations within family structures.

Conclusions:

• Gaussian mixture models with empirical priors offer a significant improvement for SNP genotype calling.
• The method enhances accuracy and reduces missing data, particularly valuable in genetic studies with family data.
• This approach provides a more robust and informative genotype calling solution for high-throughput SNP data.