Bernie Daigle
University of Memphis
Parameter estimation is a crucial step in the process of model inference for systems biology. Despite the wide variety of methods available, estimating unknown kinetic parameters is often a bottleneck for accurate computational modeling. Given experimental data and a well-defined biochemical model, several classes of methods are available for this task. These methods vary in their requirements for prior knowledge regarding parameter values as well as the information content of observed data. In this talk, I will discuss a collection of these methods for estimating parameters from well-mixed stochastic biochemical systems.
I will begin by presenting a technique for performing maximum likelihood estimation (MLE) using data observed at the single molecule level. Known as MCEM2 (Monte Carlo Expectation-Maximization with Modified Cross-Entropy Method), this method acts without requiring any prior knowledge regarding the unknown parameter values. I will then discuss three Bayesian techniques-Metropolis-within-Gibbs, particle marginal Metropolis-Hastings (PMMH), and approximate Bayesian computation (ABC)-that each require such prior knowledge. These Bayesian methods enable parameter estimation from single molecule data, noise-corrupted data, and summary statistic-transformed data, respectively. Throughout the presentation, I will compare the performance of these methods by applying each to synthetically-generated data from a canonical gene expression model.