Bayesian Tools for PBPK Models in Drug Interaction Research

描述(申请人提供)：在药物-药物相互作用(DDI)研究中，为了充分了解多种药物的药代动力学及其相互作用，基于生理学的药代动力学模型(PBPK)是研究和量化所有相互作用因素(活性部位或器官)的唯一可行手段。PBPK模型由于其复杂的结构，其模型辨识问题是众所周知的；因此，它通常借用先验生理信息来使模型可辨识。然而，传统的基于确定性PBPK模型的DDI研究忽略了PK参数的总体变化、实验噪声和先验知识中的不确定性，从而导致了可靠性未知的主观结论。这项拨款申请提出了一个贝叶斯工具系统，通过探索三个目标来应对PBPK模型的这些药理学、统计学和计算方面的挑战。在目标1中，基于已发表的动物、体外和体内数据，建立了抑制剂-底物组合的单个药物的PBPK模型；系统地讨论了从信息有限的简单模型到知识丰富的复杂模型的策略；并提出了贝叶斯荟萃分析方法。目的2，在临床DDI研究的基础上，提出了一种针对抑制剂-底物联合PBPK模型的两阶段贝叶斯方法；提出了一种预测假阴性率来评估DDI预测；并实施了贝叶斯模型选择过程以在竞争性模型中选择最佳的PBPK模型。在目标3中，所有规定的贝叶斯工具都在R中实现，这是一个统计和计算免费软件；开发其基于网络的应用程序是为了方便一般研究科学家使用它。在这项拨款申请中，我们认识到DDI是依赖酶的事实。因此，选择了一种CYP3A特异性抑制剂-底物组合，酮康唑-咪达唑仑，作为开始的例子。它将作为基于多酶和多药物的DDI PBPK模型的构建块。

DESCRIPTION (provided by applicant): In drug-drug interaction (DDI) research, to fully understand multiple drugs' pharmacokinetics and their interactions, a physiologically based pharmacokinetic model (PBPK) is the only viable means of investigating and quantifying all the interacting factors (active sites or organs). A PBPK model is known for its model identification problem because of its complex structure; hence, it usually borrows prior physiological information to make the model identifiable. However, conventional DDI researches based on deterministic PBPK models ignore population variations in PK parameters, experimental noise and uncertainties in prior knowledge, which in turn leads to subjective conclusions with unknown reliability. This grant application proposes a system of Bayesian tools to meet these pharmacological, statistical and computational challenges of PBPK models by exploring three aims. In Aim 1, individual drug's PBPK models of the inhibitor-substrate combination is established based on their published animal, in-vitro, and in-vivo data; the strategy of starting a simple model with limited information to a complex model with rich knowledge is systematically discussed; and Bayesian meta-analysis methods are proposed. In Aim 2, based on clinical DDI studies, a two-stage Bayesian method for a joint inhibitor-substrate PBPK model is developed; a predictive false negative rate is proposed to evaluate the DDI prediction; and Bayesian model selection procedures are implemented to select the best PBPK model among competitive ones. In Aim 3, all the prescribed Bayesian tools are implemented in R, a statistical and computational freeware; and its web-based application is developed to facilitate its usage for general research scientists. In this grant application, we realize the fact that the DDI is enzyme-dependent. Hence, a CYP3A specific inhibitor-substrate combination, ketoconazole-midazolam, is chosen as a starting example. It will serve as a building block for multi-enzyme and multi-drug based DDI PBPK models.