Genetic prediction in the Genetic Analysis Workshop 18 sequencing data.

Area of Science:

• Genetics and Bioinformatics
• Statistical Genomics
• Computational Biology

Background:

• High-throughput sequencing data enables phenotype prediction from genotypes, forming prognostic models.
• Extended pedigrees offer valuable relatedness information for estimating genetic parameters like heritability.
• Previous statistical models were often designed for experimental crosses, limiting flexibility for human pedigrees.

Purpose of the Study:

• To establish prognostic models for genetic prediction using high-throughput sequencing data.
• To compare standard and robust logistic regression for phenotype prediction in unrelated individuals.
• To evaluate methods for genetic prediction and model validation in extended human pedigrees.

Main Methods:

• Comparison of standard logistic regression with robust logistic regression on unrelated individuals.
• Estimation of kinship matrices and genetic values using extended family data (quantitative trait: blood pressure).
• Application of various model validation techniques, including cross-validation and within-family validation.

Main Results:

• Standard and robust logistic regression yielded similar results for phenotype prediction.
• Decorrelation approaches significantly reduced computational burden in genetic analyses.
• Some software packages exhibited issues, such as estimating negative eigenvalues for correlation matrices.

Conclusions:

• Statistical models and software require improved flexibility to accommodate diverse human pedigree structures.
• Efficient computational strategies, like decorrelation, are crucial for handling large genetic datasets.
• Further development is needed to enhance the applicability of genetic prediction models to complex human populations.