Title: Parameter Estimation for a Nonlinear HIV Dynamics Model
Speaker: Brian M. Adams, LTE Candidate, North Carolina State University
Date/Time: Wednesday, March 9, 2005, 10:00-11:00 am (PDT)
Location: Building 980, Room 95 (Sandia NM), Building 940, Room 1182 (Sandia CA)
Brief Abstract: Human Immunodeficiency Virus (HIV) currently affects 38 million people worldwide. Since discovering HIV in the early eighties, researchers have made great strides in understanding this immune system-decimating retrovirus. They have consequently implemented successful treatment strategies that delay progression to AIDS, thus prolonging a patient's life. Mathematical models of in-host infection have played key roles in these developments by helping scientists gain insight into viral dynamic mechanisms and rates.
This talk begins with an introduction to HIV infection and clinical data from a Boston-based acute HIV infection study, which motivates desirable model features. A nonlinear ODE model for HIV infection dynamics is described and a statistical/mathematical inverse problem methodology for estimating its dynamic parameters from data is presented. This methodology is motivated by varied patient outcomes over the disease time course, and therefore assumes that dynamic parameters vary between individuals. The estimates resulting from this process are therefore general probability density functions, describing the distribution of parameters across the population. In the formulation considered, the optimization problem reduces to a quadratic programming problem, albeit one that often requires regularization to achieve good results.
The proposed methods are validated on simulated data and compared to traditional nonlinear least squares methods of data fitting for obtaining individual parameter estimates. In addition to the interindividual variability captured by the parameter distribution, quantification of uncertainty in the inverse problem process is discussed, with implications on reliability of estimates from the process. Finally, parameter estimation results using viral load and T-cell count data from patients in the acute HIV infection study are discussed. It is hoped that models validated with such data can predict the efficacy of vaccines or novel treatment strategies.
CSRI POC: Scott Mitchell, (505) 845-7594 |