In Silico Assesment of Drug Metabolism and Toxicity

描述（由申请人提供）：药物开发后期分子的失败在很大程度上归因于ADME/Tox性能差。这些特性在药物发现的较早、成本较低的阶段通常是可以预测的。这项工作的目标是使用一个名为metdrug的计算套件来预测代谢和毒性。这整合了人类内源性和外源性代谢以及信号通路，也可以纳入基因表达和实验数据。在第一阶段，研究人员开发了预测主要cypp介导通路的新算法，并成功验证了预测代谢物和可能有毒的反应性代谢物的规则。该算法的开发能够预测底物和代谢物，代谢的亲和力和速率以及与其他内源性代谢和信号通路的相互作用。利用二期资金，我们将开发大型综合数据集（bbb1000个分子），用于与主要CYPs的体外药物-药物相互作用，并使用这些数据集为这些人类药物代谢酶生成机器学习算法。我们还将对大鼠和小鼠的药物代谢和这些酶的转录调控数据进行注释，捕获动力学数据，这些数据也可用于预测模型的建立。我们还将生成一种新的算法，利用第一阶段的代谢物规则来准确预测代谢物，从而为已知药物产生分子指纹。然后，已知人类代谢物分子数据库将被用作机器学习算法的输入。我们将结合各种QSAR模型对酶亲和力和代谢率的预测，这些酶的相对贡献及其组织分布，最终预测药物的清除率。这项工作将使GeneGo开发出一种独特的工具，可以改善代谢和毒性的预测。这些新功能和数据库内容随后将被推销给制药公司和学术界。

DESCRIPTION (provided by applicant): Failure of molecules in the late stages of drug development are to a large extent attributable to poor ADME/Tox properties. These properties are generally predictable in the earlier, cheaper stages of drug discovery. The goal of this work is to predict metabolism and toxicity using a computational suite called MetaDrug. This integrates human endogenous and xenobiotic metabolic as well as signalling pathways and can also incorporate gene expression, and experimental data. Under phase I, novel algorithms for predicting major CYP-mediated pathways were generated and successfully validated along with rules for predicting metabolites and reactive metabolites formed which are likely to be toxic. This algorithm development enabled the prediction of substrates and metabolites, the affinity and the rate of metabolism as well as interactions with other endogenous, metabolic and signalling pathways. With phase II funding we will develop large comprehensive datasets (>1000 molecules) for in vitro drug-drug interactions with the major CYPs, and use these for generating machine learning algorithms for these human drug metabolizing enzymes. We will also annotate rat and mouse data for drug metabolism and the transcriptional regulation of these enzymes, capturing the kinetic data which can also be used for predictive model building. We will also generate a novel algorithm for the accurate prediction of metabolites using the metabolite rules from phase I to produce a molecular fingerprint for known drugs. The database of molecules with known human metabolites will then be used as an input for a machine learning algorithm. We will combine the predictions from our various QSAR models for enzyme affinity and rate of metabolism, the relative contributions of these enzymes and their tissue distribution, to ultimately predict the clearance of a drug. The proposed work will enable GeneGo to develop a unique tool that will improve the prediction of metabolism and toxicity. These new features and database content will then be marketed to pharmaceutical companies and academia.