POPULATION PHARMACOKINETICS MODELING ON A SUPERCOMPUTER

使用潜在毒性药物进行精确治疗需要最优的方法来模拟患者的药代动力学/动力学（PK/PD）行为。在过去的两年里，我们在圣地亚哥超级计算机中心的大型并行Cray T3E机器上实现了有用的人口建模软件，创建了NCRR-USC-SDSC人口建模研究资源。这给这个计算密集型领域带来了最好和最快的硬件和软件资源。在PC上花费一周时间的工作在这个资源上只需要几分钟。在这个竞争性更新应用程序中，我们现在的目标是1)实现一种新的非参数建模算法，用于分析更大的模型；2)实现确定该非参数模型的置信限的方法，其中目前没有置信限。这将是第一次使用非参数模型，允许患者群体之间统计显著差异的方向；3)开发检测过程噪声（微分方程本身的噪声）的方法，而不是目前处理环境噪声的不正确方法，好像它是测量中的噪声，特别是在非线性模型中，并检测随时间变化的参数值（全新的东西）；4)实施更优化的方法，在最佳时间对受试者或患者的反应进行采样，避免当前的“第2条”，即为了计算获得观察该系统的样本的最佳时间，人们应该提前知道参数值；5)升级资源的用户界面，使其更加用户友好和易于使用，并实现“最大熵”关节密度，将旧的人口模型（原始原始数据消失）转换为离散关节密度，以新的“多模型”剂量设计方法进行最佳治疗。该资源现已用于克服在小型机器上进行群体PK/PD建模的瓶颈，并用于量化艾滋病联合抗病毒治疗的共同效果。它也可以用于抗癌、抗菌和其他药物治疗。为PC用户提供了一个新的类似windows的用户界面，它将更加友好和有用，可以通过互联网和万维网轻松快速地访问。

Precise therapy with potentially toxic drugs demands optimal methods to model population with pharmacokinetic/dynamic (PK/PD) behavior in patients. In the past 2 years of our current 3 year project, we have implemented useful population modeling software on the San Diego Supercomputer Center's large parallel Cray T3E machine, creating an NCRR-USC-SDSC Research Resource for Population Modeling. which brings to this computationally intensive area the best and fastest hardware and software resources available. A job taking a week on a PC takes only a few minutes on this resource. In this competing renewal application, our aims now are 1) to implement a new non-parametric modeling algorithm for analysis of still larger models; 2) to implement methods for determining confidence limits for this non-parametric models, where none exist at present. This will, for the first time with non-parametric models, permit direction of statistically significant differences between populations of patients; 3) to develop methods to detect process noise (noise in the differential equations themselves) rather than the current incorrect method of dealing with environmental noise as if it were noise in the measurements, especially in non-linear models, and to detect changing parameter values over time (something totally new); 4) to implement still more optimal methods to sample subject or patient responses at optimal times, avoiding the current "Catch 2" that one is supposed to know the parameter values in advance in order to compute the optimal times to obtain samples for observing that system; and 5) to upgrade the user interface for the resource, making it more user- friendly and easy to use, and also to implement "maximum entropy" joint densities to convert old population models (whose original raw data is gone) to the discrete joint densities needed for optimal therapy with the new "multiple model" methods of dosage design. This resource has now been used to overcome the bottleneck in population PK/PD modeling on smaller machines, and to quantify the shared effects of combinatorial antiviral therapy for AIDS. It can do the same for anti-cancer, anti-bacterial, and other drug therapy. It will be even more friendly and useful with a new windows-like user interface for the PC user, easily and rapidly accessible over the internet and the world wide web.