Ramírez-Galicia Guillermo, Garduño-Juárez Ramón, Deeb Omar, Hemmateenejad Bahram
Instituto de Ciencias Físicas, Universidad Nacional Autónoma de México, PO Box 48-3, 62250 Cuernavaca, Morelos, México.
Chem Biol Drug Des. 2008 Mar;71(3):260-70. doi: 10.1111/j.1747-0285.2008.00626.x. Epub 2008 Jan 29.
One hundred and six morphinan derivatives were taken from the Drug Evaluation Committee reports to propose several quantitative structure-activity relationship models to describe the mu-receptor-binding affinity. After several procedures to reduce the descriptor number, 21 descriptors were selected for the descriptor pool by a complete Multiple Linear Regression methodology. In this procedure only three molecules were considered as outliers. Several tests changing the relation between training:predicted sets were considered to find the best relation between these sets. The higher the number of molecules in the predicted set the higher the predictive power was observed. The optimal number of descriptors was established using the Akaike's information criterion and Kubinyi fitness function parameters. The Artificial Neuron Network methodology was applied to improve the Multiple Linear Regression best result. Finally, the regression through the origin methodology was applied to establish the best model from the Artificial Neuron Network methodology. The best quantitative structure-activity relationship model was proven to be independent of chance correlation.
从药物评估委员会报告中选取了106种吗啡喃衍生物，以提出几个定量构效关系模型来描述μ受体结合亲和力。经过几个减少描述符数量的步骤后，通过完全多元线性回归方法为描述符库选择了21个描述符。在此过程中，仅将三个分子视为异常值。考虑了几种改变训练集与预测集之间关系的测试，以找到这些集合之间的最佳关系。预测集中的分子数量越多，观察到的预测能力越高。使用赤池信息准则和库宾伊适应度函数参数确定描述符的最佳数量。应用人工神经网络方法来改进多元线性回归的最佳结果。最后，应用过原点回归方法从人工神经网络方法中建立最佳模型。最佳定量构效关系模型被证明与偶然相关性无关。
