Viña Dolores, Uriarte Eugenio, Orallo Francisco, González-Díaz Humberto
Department of Organic Chemistry, University of Santiago de Compostela, 15782, Spain.
Mol Pharm. 2009 May-Jun;6(3):825-35. doi: 10.1021/mp800102c.
There are many drugs described with very different affinity to a large number of receptors. In this work, we selected drug-receptor pairs (DRPs) of affinity/nonaffinity drugs to similar/dissimilar receptors and we represented them as a large network, which may be used to identify drugs that can act on a receptor. Computational chemistry prediction of the biological activity based on quantitative structure-activity relationships (QSAR) substantially increases the potentialities of this kind of networks avoiding time- and resource-consuming experiments. Unfortunately, most QSAR models are unspecific or predict activity against only one receptor. To solve this problem, we developed here a multitarget QSAR (mt-QSAR) classification model. Overall model classification accuracy was 72.25% (1390/1924 compounds) in training, 72.28% (459/635) in cross-validation. Outputs of this mt-QSAR model were used as inputs to construct a network. The observed network has 1735 nodes (DRPs), 1754 edges or pairs of DRPs with similar drug-target affinity (sPDRPs), and low coverage density d = 0.12%. The predicted network has 1735 DRPs, 1857 sPDRPs, and also low coverage density d = 0.12%. After an edge-to-edge comparison (chi-square = 9420.3; p < 0.005), we have demonstrated that the predicted network is significantly similar to the one observed and both have a distribution closer to exponential than to normal.
有许多药物对大量受体具有截然不同的亲和力。在这项工作中，我们选择了对相似/不同受体具有亲和力/无亲和力的药物-受体对（DRP），并将它们表示为一个大型网络，该网络可用于识别能够作用于某一受体的药物。基于定量构效关系（QSAR）的生物活性计算化学预测极大地提高了这类网络的潜力，避免了耗时耗力的实验。不幸的是，大多数QSAR模型不具有特异性，或仅预测针对一种受体的活性。为了解决这个问题，我们在此开发了一种多靶点QSAR（mt-QSAR）分类模型。在训练中，整体模型分类准确率为72.25%（1390/1924种化合物），在交叉验证中为72.28%（459/635）。该mt-QSAR模型的输出用作构建网络的输入。观察到的网络有1735个节点（DRP）、1754条边或具有相似药物-靶点亲和力的DRP对（sPDRP），覆盖密度较低，d = 0.12%。预测的网络有1735个DRP、1857个sPDRP，覆盖密度也较低，d = 0.12%。经过边对边比较（卡方 = 9420.3；p < 0.005），我们证明预测的网络与观察到的网络显著相似，且两者的分布更接近指数分布而非正态分布。
