Coalescent: an open-science framework for importance sampling in coalescent theory

Area of Science:

• Computational evolutionary biology
• Population genetics
• Statistical inference

Background:

• Coalescent theory relies on computational methods like importance sampling (IS) for statistical calculations.
• Existing computational frameworks for IS lack flexibility, user-friendliness, and robust benchmarking capabilities.
• Researchers face challenges in implementing and comparing novel IS schemes reliably.

Purpose of the Study:

• To develop a general, user-friendly Java framework for importance sampling in coalescent theory.
• To enable accurate computation of likelihoods and parameter estimation under standard coalescent models.
• To provide a platform for reliable benchmarking and comparison of different IS schemes.

Main Methods:

• Designed a Java-based framework implementing core coalescent theory concepts and IS.
• Integrated standard competing IS proposals and existing exact probability calculation frameworks.
• Incorporated an intuitive user interface, multicore processing, and cross-platform compatibility (Windows, Mac, Linux).

Main Results:

• The framework accurately computes likelihoods and maximum likelihood estimates for mutation parameters.
• Validation against established coalescent literature benchmarks confirms the framework's accuracy.
• Evaluation reveals that the efficiency order of IS schemes depends on both running time and effective sample size.

Conclusions:

• The developed framework addresses limitations of existing tools, offering flexibility and reliability for coalescent computations.
• Introduced 'just-in-time delegation' technique to optimize the balance between computational speed and precision.
• Provides a potential solution to the '2^8 programs problem' by facilitating model and scheme customization.