A parallelizable sampling method for parameter inference of large biochemical reaction models

TL;DR

This paper introduces a parallelizable sampling algorithm for efficient parameter inference in large biochemical reaction models, capable of handling high-dimensional spaces and both deterministic and stochastic systems.

Contribution

The authors present a novel, scalable algorithm that estimates model parameters from trajectory data using level set sampling and density estimation, improving inference for complex biological models.

Findings

Algorithm successfully infers parameters of large models

Effective for both deterministic and stochastic systems

Handles high-dimensional parameter spaces

Abstract

The development of mechanistic models of biological systems is a central part of Systems Biology. One major task in developing these models is the inference of the correct model parameters. Due to the size of most realistic models and their possibly complex dynamical behaviour one must usually rely on sample based methods. In this paper we present a novel algorithm that reliably estimates model parameters for deterministic as well as stochastic models from trajectory data. Our algorithm samples iteratively independent particles from the level sets of the likelihood and recovers the posterior from these level sets. The presented approach is easily parallelizable and, by utilizing density estimation through Dirichlet Process Gaussian Mixture Models, can deal with high dimensional parameter spaces. We illustrate that our algorithm is applicable to large, realistic deterministic and stochastic models and succeeds in inferring the correct posterior from a given number of observed trajectories. This algorithm presents a novel, computationally feasible approach to identify parameters of large biochemical reaction models based on sample path data.