Estimating viral infection parameters using Markov Chain Monte Carlo simulations

V. Predoi

DOI: https://doi.org/10.48550/arXiv.1501.07850

2015-01-30

Quantitative Methods

Abstract:Given a mathematical model quantifying the viral infection of pandemic influenza H1N1pdm09-H275 wild type (WT) and H1N1pdm09-H275Y mutant (MUT) strains, we describe a simple method of estimating the model's constant parameters using Monte Carlo methods. Monte Carlo parameter estimation methods present certain advantages over the bootstrapping methods previously used in such studies: the result comprises actual parameter distributions (posteriors) that can be used to compare different viral strains; the recovered parameter distributions offer an exact method to compute credible intervals (similar to the frequentist 95% parametric confidence intervals (CI)), that, in turn, using a suitable analysis statistic, will be narrower than the ones obtained from bootstrapping; given an appropriate computational parallelization, Monte Carlo methods are also faster and less computationally intensive than bootstrapping. We fit Gaussian distributions to the parameter posterior distributions and use a two-sided Kolmogorov-Smirnoff test to compare the two strains from a parametric point of view; our example result shows that the two strains are 94% different. Furthermore, based on the obtained parameter values, we estimate the reproductive number R0 for each strain and show that the infectivity of the mutant strain is larger than the wild type strain.