Parameter inference for stochastic biochemical models from perturbation experiments parallelised at the single cell...

Area of Science:

• Systems Biology
• Computational Biology
• Biophysics

Background:

• Characterizing single-cell biochemical processes requires advanced experimental and computational tools.
• Modern platforms enable high-resolution, longitudinal observation of cellular responses to individualised inputs.
• Existing parameter inference methods for stochastic models struggle with sparse data and limited perturbations.

Purpose of the Study:

• To investigate and compare computational approaches for calculating parameter likelihoods from single-cell longitudinal data.
• To focus on coupling approximations of the chemical master equation (CME) with Kalman filters.
• To address limitations of current methods when dealing with rich, high-resolution single-cell data.

Main Methods:

• Investigated parameter likelihood calculation for stochastic kinetic models using approximations of the chemical master equation (CME).
• Focused on coupling the linear noise approximation (LNA) or moment closure methods with a Kalman filter.
• Utilized filtering-based iterative likelihood evaluation for parameter inference.

Main Results:

• Coupling the LNA to a Kalman filter accurately approximates likelihoods and enables parameter inference from frequent measurements.
• This approach works even when the LNA itself poorly approximates the CME.
• Computational cost scales favorably with measurement times and input perturbations, suiting modern experimental data.

Conclusions:

• The LNA-Kalman filter coupling provides an accurate and computationally efficient method for parameter inference from high-resolution single-cell data.
• This approach is well-suited for analyzing data from advanced experimental platforms like those using optogenetics.
• Demonstrated practical utility through parameterizing a stochastic model of an optogenetic gene expression system.