Development of a model selection method for population pharmacokinetics analysis by deep-learning based reinforcement learning

For generic drug development, population pharmacokinetics (popPK) analysis is a critical part of the emerging technology of model-based bioequivalence (BE) analysis. PopPK models provide support for generalizing the conclusion of BE to groups that were not included in a BE study. The popPK model selection is essentially a multiple-objectives/variables optimization problem. Recent years have witnessed the overwhelming success of the reinforcement learning (RL) approaches in addressing optimization problem. Thus, the objective of this project is to develop a model selection method for the popPK analysis using the deep-learning based RL algorithm. Specific Aim 1: Develop a model selection method using a deep-learning based RL algorithm. A thorough survey should be conducted to gain a good understanding of the current state of the art for deep-learning based RL algorithms and their applications. The most appropriate algorithm/pipeline should be adopted to develop the model selection method. Specific Aim 2: Design simulations reflecting different scenarios of PK data, such as independent/correlated covariates, simple/complex (e.g., multiple peaks) time-concentration profiles and sparse-sampling design. The simulated datasets should be used to conduct systematic performance checks. Specific Aim 3: Identify proper metrics for performance evaluation. The selected metrics should be unbiased and mathematically/statistically meaningful. Specific Aim 4: Conduct performance evaluation. The developed model selection method and at least a stepwise regression and a genetic algorithm-based approach should be applied to the simulated datasets to perform popPK model building. The selected performance evaluation metrics should be used to compare the performance of the different methods. Specific Aim 5: Use real PK dataset(s) to demonstrate the applicability and advantage of using the developed method in popPK model building.