Imputing genotypes using regularized generalized linear regression models.

Area of Science:

• Genomics
• Bioinformatics
• Statistical Genetics

Background:

• Advancing genomic sequencing technologies enhance understanding of genotype-trait relationships.
• Missing genetic data poses a significant challenge, limiting the statistical power of genetic studies.
• Existing genotype imputation methods have limitations, particularly in selecting informative neighboring SNPs.

Purpose of the Study:

• To propose and evaluate a novel regularized generalized linear model (GLMNET) for imputing missing genotypes.
• To address limitations in previous regression-based imputation methods regarding the selection of neighboring single nucleotide polymorphisms (SNPs).
• To compare the performance of the GLMNET method against established imputation techniques.

Main Methods:

• Development of a regularized generalized linear model (GLMNET) for genotype imputation.
• Comparative analysis using two simulation scenarios: sparse-missing and small-panel expansion models.
• Evaluation on Canadian Holstein cattle and human HapMap CEU datasets, comparing GLMNET with multinomial regression, fastPHASE, and BEAGLE.

Main Results:

• The GLMNET method demonstrated superior performance in the small-panel expansion scenario.
• In the sparse-missing scenario, fastPHASE exhibited slightly better performance than GLMNET.
• Performance was assessed across diverse datasets, including livestock and human populations.

Conclusions:

• The GLMNET approach offers a promising alternative for genotype imputation, especially in specific data scenarios.
• The study highlights the trade-offs between different imputation methods depending on the missing data patterns.
• Accurate genotype imputation is crucial for maximizing the utility of genomic data in research.