Speaker
Description
Mathematical models are invaluable for understanding and predicting how biological systems behave. Constructing such models requires specifying the mechanisms and relationships involved, which in practice are not perfectly known. In the presence of multiple competing models, principled approaches to modeling need to account for model uncertainty. Bayesian model averaging (BMA) provides a framework for testing mechanistic hypotheses and generating predictions under model uncertainty. BMA requires the calculation of model evidence, which may be obtained via deterministic approximations or estimated using Monte Carlo methods. On the other hand, there is a growing recognition that parameter non-identifiability—the inability to distinguish between parameter values given available data—is pervasive in mathematical biology and has important consequences for statistical inference. In this work, we investigate the reliability of evidence computation methods when parameter non-identifiability is present, and find that deterministic approximations can produce misleading model selection results, as their underlying assumptions are violated. We propose an efficient variant of adaptive multiple importance sampling for evidence estimation, and demonstrate its robustness against non-identifiability. We use ecological case studies to show that our methods yield BMA results that are comparable to those obtained by Markov chain Monte Carlo methods at substantially lower computational cost.
