Simulation-Based Methods for Optimal Sampling Design in Systems Biology

In many research areas of systems biology including virology, pharmacokinetics, and population biology, researchers often use the dynamical systems to describe dynamics of biological systems. We can learn about these biological systems by estimating the parameters of the dynamical systems from sampling data. Therefore, an important question arises: how can one select the optimal sampling points for accurately estimating the parameters? Classical methods often rely on Fisher information matrix-based criteria such as A-, D-, and E-optimality. However, these methods require an initial estimate of the parameters and often lead to a suboptimal result when the initial estimate is inaccurate. In this paper, we develop two simulation-based methods for optimal sampling design that do not require an initial estimate of the parameters. The first method, E-optimal-ranking (EOR), employs the E-optimal criterion while the second method utilizes the Long short-term memory (LSTM) neural network. We demonstrate the performance of our proposed methods using simulation studies based on two popular models in system biology (Lotka-Volterra and Three-compartment models). The results show that our methods outperform the naive random selection method and the classical E-optimal design.