Computationally Efficient Composite Likelihood Statistics for Demographic Inference.

Area of Science:

• Population Genetics
• Computational Biology
• Statistical Inference

Background:

• Population genetics tools often use composite likelihoods due to the difficulty of modeling genomic linkage.
• Estimating parameter uncertainties and model selection traditionally require full likelihoods, necessitating computationally intensive maximum-likelihood estimation (MLE) on bootstrapped data.

Purpose of the Study:

• To develop a computationally efficient method for statistical inference in population genetics.
• To adapt existing demographic inference tools for faster and robust analysis using adjusted composite likelihoods.

Main Methods:

• Applied statistical theory to adjust composite likelihoods within population genetics software.
• Validated the adjusted composite likelihood approach in two demographic inference tools: ∂a∂i and TRACTS.
• Compared the performance of adjusted composite likelihoods against traditional MLE bootstrapping on simulated and real genomic data.

Main Results:

• The adjusted composite likelihood method provides statistically robust inference.
• Performance of the adjusted method is comparable to MLE bootstrapping in accuracy.
• Achieved significant computational time savings, using orders of magnitude less time than MLE bootstrapping.

Conclusions:

• Adjusted composite likelihoods offer a computationally efficient alternative for statistical inference in population genetics.
• This approach enhances the usability of tools like ∂a∂i and TRACTS for demographic inference.
• The findings pave the way for faster and more scalable analyses of genomic data.