Least squares estimation of variance components for linkage

Area of Science:

• Statistical genetics
• Quantitative genetics
• Bioinformatics

Background:

• Variance component estimation is crucial for genetic linkage studies.
• Maximum likelihood (ML) methods are computationally intensive.
• Efficient estimation methods are needed for complex genetic data.

Purpose of the Study:

• To develop and evaluate least squares (LS) procedures for variance component estimation.
• To compare the efficiency and speed of LS versus ML methods.
• To assess the performance of LS across different data distributions (normal, bivariate normal, skewed).

Main Methods:

• Development of simple least squares (LS) expressions for variance component estimation.
• Comparative simulations of LS and maximum likelihood (ML) procedures.
• Evaluation of estimator bias and efficiency for normal, bivariate normal, and skewed data.
• Computational performance analysis comparing LS and ML speed.

Main Results:

• LS procedures are computationally rapid, over 4,000 times faster than ML for bivariate data.
• For normal data, LS estimators are unbiased but less efficient (<50%) than ML.
• For bivariate normal data, LS efficiency relative to ML improves (generally >60%).
• LS demonstrates markedly higher efficiency than ML for parameter estimation with skewed data.

Conclusions:

• LS methods provide a computationally efficient alternative for variance component estimation.
• LS is particularly advantageous for skewed or bivariate genetic data.
• LS methods are recommended for initial interval mapping in multivariate genetic studies due to their speed.