Graduate Student LaMont Cannon will be defending his dissertation on Thursday, December 7th at 9:15 – 10:15 am in 5 Innovative Way at Delaware Technology Park.
Date: Thursday, December 7
Time: 9:15-11:15 AM
Location: 5 Innovative Way Conference room (Delaware Technology Park)
Committee Chair: Ryan Zurakowski
Committee: Abhyudai Singh, Nii Attoh-Okine, Fabrizio Sergi
Information Optimal Experiment Design of HIV 2-LTR Clinical Trials By Expected Kullback-Leibler Divergence
Finding a cure for individuals infected with the Human Immunodeficiency Virus (HIV) has proved to be a challenging task. This is primarily due to the fact that conventional treatment has not been able to adequately disrupt the replication process in order to eradicate the virus. One of the possible explanations for this lack of treatment efficacy is that there are low levels of ongoing replication occurring in locations of reduced drug concentration called sanctuary cites. In order to effectively treat the disease, it would be advantageous to clinicians to know how much on-going replication is occurring. This knowledge would then help to guide patient specific treatment for the disease. A novel method to quantify the level on going replication has been suggested. In order to be identified as a valid method clinical trials must be carried out; however, they can often be costly, time consuming and demanding to the patients. For these reasons, meticulous effort should be applied to make sure that these trials are as efficient and informative as possible
This thesis summarizes several common methods used for optimal design that can be used to address these issues. A mathematical model is employed to demonstrate the dynamics of the system. Using this model in conjunction with preliminary laboratory data, Markov Chain Monte Carlo Methods are applied to estimate model parameter distributions under a variety of different experiment assumptions. We then calculate the Expected Kullback-Leibler Divergence (EKLD) between the a priori parameter distributions and the a posteriori distributions for each experiment regimen. This value is taken to indicate the amount of information we can expect to gain from performing the experiment under that particular design. Through the use of genetic algorithms we then locate the experiment design that optimized the expected gain in information. In doing so, this thesis shows that the EKLD optimization method is robust and performs equally well if not better than traditional optimal experiment design techniques under multiple experiment design criteria.