Likelihood-free nested sampling for parameter inference of biochemical reaction networks

Area of Science:

• Systems Biology
• Computational Biology
• Statistical Inference

Background:

• Accurate parameter inference is crucial for developing mechanistic models in Systems Biology.
• Nested sampling methods offer parallelization and error estimates but require tractable likelihood functions.
• Stochastic models often feature intractable likelihoods, limiting the application of standard nested sampling.

Purpose of the Study:

• To develop a likelihood-free nested sampling method for parameter inference in Systems Biology.
• To enable the analysis of complex stochastic systems with intractable likelihoods.
• To provide an unbiased estimator of Bayesian evidence and posterior samples.

Main Methods:

• Developed a likelihood-free nested sampling algorithm.
• Derived a lower bound on the estimator's variance to formulate a novel termination criterion.
• Applied the method to realistically sized models with simulated and biological data.
• Compared the new method with pMCMC and ABC-SMC.

Main Results:

• The likelihood-free nested sampling method provides an unbiased estimator of Bayesian evidence and posterior samples.
• The method successfully infers parameters for complex stochastic systems.
• A novel termination criterion based on variance estimation improves reliability.
• Performance was validated against established likelihood-free approaches.

Conclusions:

• The presented likelihood-free nested sampling method significantly advances parameter inference for complex biological systems.
• This approach overcomes the limitations of traditional methods for models with intractable likelihoods.
• The method offers reliable posterior inference and variance estimation for stochastic systems.