Statistical confidence for likelihood-based paternity inference in natural populations

Area of Science:

• Population genetics
• Molecular ecology
• Conservation genetics

Background:

• Paternity inference in natural populations commonly uses polymorphic codominant markers.
• Existing methods struggle with incomplete sampling and data errors typical in large-scale studies.
• Interpreting paternity likelihoods from non-excluded males remains challenging.

Purpose of the Study:

• To develop a robust method for paternity inference using codominant markers.
• To address challenges of typing error, incomplete sampling, and data gaps in natural populations.
• To provide a statistically sound framework for assigning paternity with known confidence.

Main Methods:

• Derivation of likelihood ratios for paternity inference incorporating typing error.
• Definition of a statistic 'delta' for resolving paternity.
• Development of a simulation program using population-specific allele frequencies to establish confidence criteria for delta.
• Accounting for candidate male numbers, sampling proportions, and genetic data quality.

Main Results:

• The new method accurately assigns paternity with known statistical confidence, even with data errors and incomplete sampling.
• The 'delta' statistic and simulation criteria effectively resolve paternity.
• The method demonstrates robustness to the presence of relatives under common conditions.
• Successful application to the red deer (Cervus elaphus) population on the island of Rum, Scotland.

Conclusions:

• The developed method provides a reliable approach for paternity inference in natural populations using codominant markers.
• The CERVUS software facilitates allele frequency calculation, simulations, and parentage analysis.
• This advancement improves the accuracy and statistical rigor of genetic parentage studies in wildlife research.