Analysis of HIV-1 compartmental model parameters using Bayesian MCMC estimation

Abstract

In previous work Bayesian Markov-Chain Monte Carlo techniques were used to identify HIV dynamic parameters from patient data. This study used viral load data from HIV patients during interrupted cycles of antiretroviral drugs. The data were fit to a well-established single-compartment model of HIV dynamics. Experimental evidence supports the use of a new compartmental model that includes the re-circulation of T-cells between anatomical reservoirs. If the infection dynamics are indeed compartmentalized, it is not clear how this would affect the estimated values of the dynamic parameters. In this study, we identify parameters of the simple, one-compartment model using data generated by the complex spatial model, and investigate the bias in parameter values introduced by this method. Multiple instances of simulated noisy data was generated from the spatial model using parameter estimates from previous studies. Markov-Chain Monte Carlo methods were used to identify parameter values for the simple model from the simulated data, and the identified values were compared with the true values to determine the existence of bias. The maximum likelihood bias in the median estimates of the proliferation, death and infection rate parameters for target T-cells and virion production rates were on average smaller than one standard deviation of the reported parameter uncertainties for the simple model. The median estimates of the death rate of infected cells and the efficacy of the drug exhibited an average positive bias (the posteriors were larger than the priors) that was larger than one-standard deviation of the prior for all patients. Neglecting the spatial dynamics does not seem to significantly affect the estimation of the proliferation, death, and infection rate parameters for target T-Cells. Conversely, the values of the infected cell death rates and drug efficacies exhibited a consistent bias when estimated using the simplified model. Neglecting spatial dynamics will result in a consistent overestimation of the values of these parameters.