Abstract

Models are applied widely in the biological sciences and elsewhere but their usefulness depends on the quality of calibration to observed data. Stochastic individual-based models are becoming more popular but their calibration is beset by two particular difficulties: likelihood intractability and high computational expense. To overcome both these limitations, we develop a novel adaptive emulation-based calibration algorithm. We apply the algorithm to a deterministic HPV -16 model and compare the results with full MCMC before extending its application to a stochastic individual-based model of HPV -16. It works well in both cases, fitting the model to the observed data and inferring similar parameter values as the full MCMC analysis (but with lower variance). It also provides similar estimates of the mean and variance values of inferred parameters for both the deterministic and stochastic models.