Bayesian Estimation for Stochastic Gene Expression Using Multifidelity Models.

Area of Science:

• Computational biology
• Systems biology
• Biophysics

Background:

• The finite state projection (FSP) approach is crucial for inferring stochastic models of single-cell gene regulation dynamics.
• However, FSP is computationally intensive, hindering parameter inference and uncertainty quantification for complex models.

Purpose of the Study:

• To develop computationally efficient methods for Bayesian inference of stochastic gene expression parameters from single-cell data.
• To address the computational limitations of the FSP approach in analyzing complex gene regulatory networks.

Main Methods:

• Formulation and verification of an adaptive delayed acceptance Metropolis-Hastings (ADAMH) algorithm using reduced Krylov-basis projections of the FSP.
• Introduction of a hybrid ADAMH scheme combining model reduction and efficient posterior sampling.
• Comparison of ADAMH variants against a standard adaptive Metropolis algorithm with full FSP likelihood evaluations using simulated data.

Main Results:

• The proposed ADAMH variants achieve substantial computational speedup compared to the full FSP approach.
• Demonstrated efficiency on three example gene regulation models.
• Validation through comparison with existing adaptive Metropolis algorithms.

Conclusions:

• The developed ADAMH algorithms significantly reduce the computational cost of parameter estimation in stochastic gene expression models.
• These methods are expected to facilitate efficient data-driven analysis of more complex gene regulatory models.
• Enables broader application of computational modeling in single-cell biology.