遗传算法对QSAR研究中变量选择的应用

APPLICATION OF GENETIC ALGORITHMS TO VARIABLE SELECTION IN QSAR STUDIES

将遗传算法引入定量结构活性关系(QSAR)研究中,对变量进行选择,可同时建立几种比较好的QSAR模型,并以交互验证的决定系数作为适应函数,以保证模型质量的可靠性.将其分别应用于氯代酚和单取代苯系列化合物,均得到较好的结果.同时,在这些应用中也反映了遗传算法在变量选择中存在的局限性,即不能保证选择的所有变量在模型中都有显著贡献.

Genetic algorithms are applied to variable selection in quantitative structure-activity relationship (QSAR) studies. The method can build multiple better QSAR models in one run. The determination coefficient of cross-validation is introduced as the fitness function, which guarantees the predictive ability of the models. The results obtained in application to chlorophenols and monosubstituted benzene derivatives are better than class regression results. But, in the variable selection is a limitation , i.e. not all the selection variables have a significant contribution in the models.