Maximizing Information from Noisy Measurements of Single-cell Gene Expression Distributions

Appropriate experiment design is paramount to the success of model-based prediction and control of single-cell gene expression. Because of the inherent randomness of biochemical reactions, the number of measurements and sampling times need to be considered carefully to produce data that adequately constrain model parameters. Random measurement error, however, is a complicating factor that could reduce or distort the information contained in an experiment. We propose a conceptual framework based on the Fisher Information Matrix and probability kernel together with a computational method based on the finite state projection algorithm to systematically study the impact of single-cell measurement noise on stochastic model parameter identifiability. Our approach is versatile and is particularly well-suited for studying single-molecule experiments on gene expression systems where molecular species have low copy numbers and where measurement noise distribution assumes a non-symmetric, non-Gaussian shape.