Matching models of HIV-1 viral dynamics to clinical data

Creating individualized within-host multiple phase disease models of HIV-1 infection has long been a goal of mathematicians and biologists. The challenge is in trying to build models that are representative of the disease, include realistic parameter estimates, and are able to incorporate a changing model structure. In this paper, we propose a fitting procedure, motivated by the biology of the disease, for matching parameters of differential equation and stochastic models of HIV-1 infection to data, which leverages high performance computing resources. The search uses knowledge of the biological set points to restrict the search domain, and parallel simulated annealing to match the model to acute and early chronic phase patient data. We highlight this method by finding parameters for two interconnected models of HIV-1 infection which we have developed. The high quality of our data allows us to model not only viral data, but also CD4 count data through the acute and chronic phases of the disease. The time span of our model exceeds that of previous models. The algorithm is able to find parameter values for four patients consistent with literature ranges and display individual set point equilibration and disease progression for both clinical markers.