Description
Human immunodeficiency virus (HIV) is a retrovirus that compromises a person’s immune system by diminishing the number of CD4+ T-cells available to fight foreign bodies. Despite successes of antiretroviral therapies on suppressing the viral load in HIV patients, these treatments are unable to eradicate the virus. The primary obstacle to HIV eradication is resting CD4+ T-cell reservoirs, which may remain dormant for a long time and often cause viral rebounds after treatment cessation. Many mathematical models have been proposed to explore the behavior of the latent reservoir but much remains uncertain on the accurate prediction of the frequency of latently infected cells, mainly because of limited data due to complexity of measuring such cells. In this thesis, we develop a hybrid approach with mathematical modeling and machine learning combined to predict the frequency of resting CD4+ T-cell infection (RCI) reservoirs. Our framework takes advantage of the insights obtained from mathematical modeling to generate large number of reliable data sets that are then used in machine learning applications to predict latently infected cells. We found that the predictions based on our framework are consistent with the experimental data. We identified that the RCI frequency measurements with the lag time of four months can better predict the RCI frequency. Such information can be helpful for medical examiners in determining a reasonable interval between measurements of RCI frequency. Our hybrid approach can be widely applicable to many other cases in which the data is limited due to the requirement of a stringent assay for their measurement.
