基于非线性构效关系的数据发掘在药物发现中的应用

数据发掘和定量的构效关系(QSAR)是药物发现中重要的研究方法．本首先介绍了一种极具实用价值的无模式分类与模式识别的基本方法，即k最近邻法．该法应用于QSAR研究时通过若干电子、拓扑结构参数计算分子间的欧氏距离来比较其相似性，并以最相似若干分子已知活性的加权平均来估算新结构的活性数据．整个运算过程利用模拟退火法并结合遗传算法和留一-交叉验证法优选出适当的结构参数和近邻数目．本以一组已知活性的单核苷酸分子为例，对NCI和Maybridge化学数据库的分子运用kNN法进行数据发掘，通过计算数据库中分子与已知活性分子的相似性来进行筛选和活性预测，从而得到与已知结构最相似的分子，为搜寻潜在前导化合物提供重要参考信息．

DATA MINING BY NONLINEAR QSAR MODELS FOR DRUG DISCOVERY

Data mining and quantitative structure-activity relationship (QSAR) are fundamental approaches for drug discoverty research. Effective and expeditious tools would be undoubted options for searching and optimizing the lead compound. A nonlinear modeling approach, k-Nearest-Neighbor, for QSAR modeling is introduced; furthermore, a sample case is also provided by employing this method to perform data mining studies on two public databases, NCI and Maybridge, in order to find potential lead compounds from a group of mononucleotides.