Predicting the Volume of Distribution of Drugs and Toxicants with Data Mining Methods

世纪初的一位医生帕拉塞尔苏斯（Paracelsus）有这样一句话：“所有的东西都是毒药，没有什么是无毒的;只有剂量才能让某些东西无毒”（http：//en.wikipedia.org/wiki/Paracelsus）。尽管在过去几十年中药理学取得了重大进展，但目前仍然很难找到药物应该给予患者多少，多久和多长时间的问题的好答案，以最大限度地提高其治疗效果并尽量减少其不良反应。这些问题是药代动力学和药效学相关领域关注的中心问题。药代动力学涉及药物如何被身体加工，即，药物输入参数（例如剂量中的药物量和给药频率）与体内药物浓度随时间的关系。相比之下，药效学关注的是药物如何影响身体，即，药物浓度与药物治疗和副作用随时间的关系。该项目侧重于一个重要的药代动力学问题：如何估计药物的分布容积，它代表药物一旦进入全身进入体内后分布的体积。药物分布容积的估计是非常重要的，因为它可以预测药物在体内的血药浓度，并影响药物的半衰期，这反过来又是非常重要的，以确定正确的剂量方案，临床医生应该处方给病人。本项目旨在开发新的计算数据挖掘方法来预测药物的分布容积。该项目的数据挖掘上下文是回归任务，其中系统被赋予一组代表一组对象的实例，其中每个实例由一个目标（响应）属性（或因变量）和一组描述对象的预测器属性（特征或自变量）组成。然后，系统发现一个回归模型，该模型基于实例的预测器属性的值来预测该实例的目标属性的值。在本项目中，待分类的对象是化合物或医疗药物，待预测的目标属性是药物的分布体积，预测属性是指药物的几种分子和物理化学性质。将在该项目中开发的数据挖掘方法与用于预测药物分布量的传统数据分析方法进行比较。

Paracelsus, a physician in the early 16th century, is credited with the phrase: "All things are poison, and nothing is without poison; only the dose permits something not to be poisonous" (http://en.wikipedia.org/wiki/Paracelsus). Despite significant advances in pharmacology in the last decades, at present it is still very difficult to find good answers to the questions of how much, how often and for how long a drug should be given to a patient, in order to maximize its therapeutic effect and minimize its adverse effects. These problems are the central concern of the related areas of pharmacokinetics and pharmacodynamics. Pharmacokinetics is concerned with how a drug is processed by the body, i.e., the relationship between drug input parameters (e.g. amount of drug in a dose and dose frequency) and the concentration of the drug in the body with time. In contrast, pharmacodynamics is concerned with how a drug affects the body, i.e., the relationship between drug concentration and the therapeutic and adverse effects of the drug with time.This project focuses on an important pharmacokinetics problem: how to estimate the volume of distribution of a drug, which represents the volume into which a drug is distributed once it has entered systemically into the body. Estimating a drug's volume of distribution is important because it predicts the drug's plasma concentration for a given amount of drug in the body and it influences the drug's half-life, which in turn is very important to determine the correct dosage regimen that clinicians should prescribe to patients.This project aims at developing new computational data mining methods to predict the volume of distribution of drugs. The data mining context for this project is the regression task, where the system is given a set of instances representing a set of objects, where each instance consists of a target (response) attribute (or dependent variable) and a set of predictor attributes (features or independent variables) describing an object. Then the system discovers a regression model that predicts the value of the target attribute for an instance based on the values of its predictor attributes. In this project, the objects to be classified will be chemical compounds or medical drugs, the target attribute to be predicted will be a drug's volume of distribution and the predictor attributes will refer to several types of molecular and physicochemical properties of drugs. The data mining methods to be developed in the project will be compared against traditional data analysis methods used for predicting a drug's volume of distribution.