Tools for Prediction of ADME-Tox Properties

该项目是 CADD 小组与国防部 (DoD) 多个小组联合项目的一部分，项目名称为“将小分子转化为研究性新药的计算平台”。国防部方面的项目牵头 PI 是 S. Anders Wallqvist 博士，他来自部队健康防护三军生物技术高性能计算软件应用研究所 (BHSAI)、远程医疗和先进技术研究中心 (TATRC)、美国陆军医学研究和物资司令部 (USAMRMC)，地址：2405 Whittier Drive, Suite 200, Frederick, MD 217602。沃尔特·里德陆军研究所 (WRAIR) 生物化学系以及美国陆军传染病医学研究所 (USAMRIID) 细胞生物学和生物化学系。整个项目的目标是整合临床前药物开发阶段的三个基本方面，即基于结构的药物设计、药理学数据的分析和预测，以及从化学结构预测不良和脱靶效应，特别是与药物代谢相关的效应。 Pugliese 博士工作中最重要的方面涉及新陈代谢和代谢物。这项工作于 2010 年初有效开始，前 CADD 小组成员 Pugliese 博士致力于实施一种资源，以成功预测类药物小分子的代谢和代谢物，作为我们计算机辅助药物设计能力的一部分，直到他于 2011 年 6 月离开 NCI 并获得永久职位。虽然这些资源的初步测试和应用是在国防部感兴趣的病原体背景下进行的，但预测代谢稳定性、代谢概况和特定代谢物的一般能力小分子适用于所有类型的药物开发，因此对于开发针对 NCI 高度感兴趣的分子靶点的抗癌疗法以及 NCI 的抗 HIV 药物设计项目非常有用。因此，即使在 2011 年夏季完成与国防部小组的正式合作后，该项目仍在继续。该项目的第一阶段已成功完成，包括在预测计算机工具领域进行调查以及可用于测试这些工具和开发（更好）预测模型的数据集。商业和免费资源均已编译或获取。已完成、提交并出版了一项已发表适当出版物的比较和基准研究。在该项目的这一部分中，我们重点关注（预测）代谢稳定性数据，例如人肝微粒体或人肝细胞测定中的半衰期值。本文还包括一项关于细胞色素 P450 相互作用（底物、抑制剂和诱导剂）预测的小型基准研究。在该项目的第二个更广泛应用的阶段，我们基于在人肝微粒体中测量的体外半衰期测定数据，开发了化合物代谢稳定性的 QSAR 模型。使用开源和商业程序（KNIME、GUSAR、StarDrop）中实现的不同统计方法和描述符集生成了各种 QSAR 模型。使用来自公共和商业数据源的四个不同的外部验证集对获得的模型进行比较，其中包括两个较小的人体体内半衰期数据集。最具预测性的模型用于预测开放 NCI 数据库中化合物的代谢稳定性，其结果已在 NCI/CADD Group 网络服务器 (http://cactus.nci.nih.gov) 上公开发布。这项研究和上述论文均发表在《Future Medicinal Chemistry》杂志上。目前的工作重点是将我们的预测能力扩展到小分子吸收、分布、代谢、排泄和毒性 (ADME/Tox) 领域所有类型特性的模型。最近，Alexey Zakharov 博士在 NCI/CADD Group 网络服务器上以化学活性预测器 (CAP) 网络服务的形式提供了一套物理化学性质、毒性以及一些生物活性的预测模型。在此主题的背景下，CADD 小组成员还开发和改进了与 QSAR 相关的通用方法和算法，并分析了 (Q)SAR 模型对来自特定项目和大型公共数据库（例如 PubChem 和 ChEMBL）的检测数据的混合匹配能力的依赖性。我们与俄罗斯同事一起对 Q(SAR) 数据和方法进行了进一步分析，其中包括几篇“混合匹配”问题分析的论文。此外，正在对 SAVI 项目（项目 6）的分子进行大型 ADME-Tox 计算。

This project was started as part of a joint project of the CADD Group with several groups at the Department of Defense (DoD), with the title Computational platforms for transforming small molecules into investigational new drugs. The projects lead PI on the DoD side was Dr. S. Anders Wallqvist, Tri-Service Biotechnology High-Performance Computing Software Applications Institute for Force Health Protection (BHSAI), Telemedicine and Advanced Technology Research Center (TATRC), U.S. Army Medical Research and Materiel Command (USAMRMC), 2405 Whittier Drive, Suite 200, Frederick, MD 217602. Other participating groups were at the Department of Biochemistry, Walter Reed Army Institute of Research (WRAIR), and the Department of Cell Biology and Biochemistry, U.S. Army Medical Research Institute for Infectious Diseases (USAMRIID). The aim of the overall project was to integrate three fundamental aspects of the preclinical drug development phase, i.e., structure-based drug design, analysis and prediction of pharmacological data, and the prediction of adverse and off-target effects, in particular those related to drug metabolization, from chemical structures. The most important aspect of Dr. Pugliese's work concerned metabolism and metabolites. The work having effectively started in early 2010, the former CADD Group member Dr. Pugliese worked on implementing a resource for successful prediction of metabolism and metabolites of drug-like small molecules as part of our computer-aided drug design capabilities, until his departure from NCI for a permanent position in June, 2011. While the initial tests and application of these resources were done in the context of pathogens of interest to DoD, the general capability of predicting metabolic stability, metabolization profile and specific metabolites of a small molecule is applicable to all types of drug development, and therefore is very useful in the development of anti-cancer therapeutics aiming at molecular targets of high interest to NCI, as well as in, e.g., NCI's anti-HIV drug design projects. The project has therefore been continued even after the completion of the formal collaboration with the DoD groups in summer 2011. The first phase of this project, consisting of canvassing the field for predictive computer tools as well as data sets that can be used to test these tools and develop (better) predictive models, has been successfully completed. Both commercial and free resources have been compiled or acquired. A comparison and benchmark study with appropriate publication was completed, submitted, and is in press. In this part of the project, we focused on (prediction of) metabolic stability data such as half-life values in Human Liver Microsome or Human Hepatocyte assays. This paper also includes a small benchmark study of predictions of cytochrome P450 interactions (substrates, inhibitors, and inducers). In the second, more-applied, phase of the project, we developed QSAR models for metabolic stability of compounds, based on in vitro half-life assay data measured in human liver microsomes. A variety of QSAR models were generated using different statistical methods and descriptor sets implemented in both open-source and commercial programs (KNIME, GUSAR, StarDrop). The models obtained were compared using four different external validation sets from public and commercial data sources, including two smaller sets of in vivo half-life data in humans. The most predictive models were used for predicting the metabolic stability of compounds from the Open NCI Database, the results of which have been made publicly available on the NCI/CADD Group web server (http://cactus.nci.nih.gov). Both this study and the paper mentioned above have been published in the journal Future Medicinal Chemistry. Current efforts focus on broadening our predictive capabilities to models of all types of properties in the area of absorption, distribution, metabolism, excretion, and toxicities (ADME/Tox) of small molecules. Recently, Dr. Alexey Zakharov has made available a suite of predictive models for physicochemical properties, toxicities, as well as some biological activities in the form of the Chemical Activity Predictor (CAP) web service on the NCI/CADD Group web server. In the context of this topic, CADD Group members have also developed and improved general QSAR-related approaches and algorithms, as well as analyzed (Q)SAR models' dependency on mix-and-match'ability of assay data coming both from specific projects and large public databases such as PubChem and ChEMBL. Further analyses of Q(SAR) data and approaches have been performed with our Russian colleagues, which includes several papers of "mix-and-match" issue analyses. Also, large ADME-Tox computations are being performed for molecules from the SAVI project (Project 6).