紫杉醇类似物抗癌活性构效关系的神经网络模式识别研究

NEURAL NETWORK PATTERN RECOGNITION STUDY ON THE QUANTITATIVE STRUCTURE-ACTIVITY RELATIONSHIPS OF TAXOL ANALOGUE ANTI-TUMOR DRUGS

经对MM3、MM+、MNDO、PM3几何优化结果进行比较,选用速度最快且精确度较好的MM3分子力学方法计算43个紫杉醇类似物的优势构型,应用MNDO法计算了化合物的电子结构,并用回归分析和BP神经网络模式识别方法寻找其电子结构与抗癌活性的关系.结果表明:(1)紫杉醇类似物及C13侧链的油水分配系数与活性参数间呈抛物线关系,最适油水分配系数Popt=3.14;(2)2-OB z中B z基团的负电荷密度越大,C1、C3原子的正电荷密度越大对活性越有利;(3)R1、R2、1-OH和2-OB z基团可能是药物与受体作用的重要部位.四参数的定量构效关系显著性较好,神经网络模式识别总识别率为98%,可较精确地预测化合物的抗癌活性.

After the MM3 geometry optimizating, the electronic structures of 43 taxol analogues have been calculated by MNDO quantum chemical method. The BP artificial neural network pattern recognition has been performed for analyzing the QSAR of taxol analogue anti - tumor drugs. It can be concluded as follows： （ 1 ） The relationship between logRr values of C13 side chain in taxol analogues and activity follows the parabola model （Popt =3.14）. （2） The more negative charge of Bz on 2 - OBz and positive charge of C1 and C3＇ atoms of taxol analogues, the higher activities they will have. （3） R1, R2, 1 -OH and 2 -OBz may be the important active sites of the molecules. The BP artificial neural network pattern recognition has a 98% accuracy, which may predict the activities of taxol analogues.