A conditional likelihood is required to estimate the selection coefficient in ancient DNA.

Area of Science:

• Population genetics
• Evolutionary biology
• Genomics

Background:

• Time-series of allele frequencies are crucial for understanding natural selection.
• Traditional methods for estimating selection coefficients rely on comprehensive fitness landscape data.
• Limited data, particularly from ancient DNA, poses challenges for accurate estimation.

Purpose of the Study:

• To develop a novel likelihood function for estimating selection coefficients from single allele frequency trajectories.
• To address the statistical limitations of using incomplete fitness landscape data.
• To improve the accuracy of selection coefficient estimation in scenarios with limited data.

Main Methods:

• Utilized two population genetics models.
• Developed a conditional likelihood function tailored for single trajectories.
• Compared the performance of the conditional likelihood with the traditional unconditioned likelihood.

Main Results:

• The conditional likelihood function accurately estimates selection coefficients even with limited fitness landscape coverage.
• This method performs well when allele frequencies are near fixation, where traditional methods fail.
• Demonstrated the limitations of unconditioned likelihoods, which can provide unfalsifiable and incorrect estimates.

Conclusions:

• The proposed conditional likelihood method offers a more precise estimation of selection coefficients from single allele frequency trajectories.
• This approach is particularly valuable for analyzing ancient DNA data with limited trajectories.
• Highlights the need for specialized methods when dealing with constrained evolutionary data.