Faculty of Pharmaceutical Sciences, M. D. University, Rohtak, India.
Arch Pharm (Weinheim). 2012 Dec;345(12):989-1000. doi: 10.1002/ardp.201100316. Epub 2012 Sep 4.
In the present study both classification and correlation techniques of diverse nature were successfully employed for the development of models for the prediction of human immunodeficiency virus (HIV) integrase inhibitory activity using a dataset comprising 50 analogs of quinolone carboxylic acid. The values of various molecular descriptors (MDs) for each analog in the dataset were computed using the MDS V-life science QSAR plus module. The values of other MDs which are not part of MDS V-life science were computed using an in-house computer program. A decision tree (DT) was constructed for the HIV integrase inhibitory activity to determine the importance of MDs. The DT learned the information from the input data with an accuracy of 98% and correctly predicted the cross-validated (10 fold) data with an accuracy of 96%. Three MDs, E-state contribution descriptor (SssOHE), molecular connectivity topochemical index ($\chi {}^{{\rm A}} $), and eccentric connectivity topochemical index ($\xi {{\rm C}}^{{\rm C}} $), were used to develop the models using moving average analysis (MAA). The accuracy of classification of single descriptor based models using MAA was found to vary from a minimum of 96% to a maximum of 98%. The statistical significance of the models was assessed through specificity, sensitivity, overall accuracy, Mathew's correlation coefficient, and intercorrelation analysis. The widely used methods like multiple linear regression, partial least squares, and principal component regression were employed for development of correlation models. The models were generated on a training set of 36 molecules. The models had a correlation coefficient (r(2) ) of 0.86 to 0.92, significant cross validated correlation coefficient (q(2) ) of 0.79 to 0.85, F-test from 63.2 to 93.06, r(2) for external test set (pred_r(2) ) from 0.69, coefficient of correlation of predicted dataset (pred r(2) Se) of 0.77, and degree of freedom from 27 to 30. Alignment independent descriptors, SsOHE-index, SaaCHE index, SssCH2, and x log P were found to be the most important descriptors for the development of correlation models for the prediction of HIV integrase inhibitory activity.
在本研究中，成功地将不同性质的分类和相关技术应用于使用包含 50 种喹诺酮羧酸类似物的数据集来开发用于预测人类免疫缺陷病毒 (HIV) 整合酶抑制活性的模型。使用 MDS V-life science QSAR plus 模块计算数据集每个类似物的各种分子描述符 (MD) 值。不属于 MDS V-life science 的其他 MD 值使用内部计算机程序计算。为 HIV 整合酶抑制活性构建决策树 (DT)，以确定 MD 的重要性。DT 使用 98%的准确性从输入数据中学习信息，并使用 96%的准确性正确预测交叉验证 (10 倍) 数据。三个 MD，E-state 贡献描述符 (SssOHE)、分子连接拓扑指数 ($\chi {}^{{\rm A}} $) 和偏心连接拓扑指数 ($\xi {{\rm C}}^{{\rm C}} $)，用于使用移动平均分析 (MAA) 开发模型。发现使用 MAA 的基于单个描述符的模型的分类准确性从最低的 96%到最高的 98%不等。通过特异性、敏感性、总体准确性、马修相关系数和互相关分析评估模型的统计学意义。使用广泛使用的方法，如多元线性回归、偏最小二乘和主成分回归，用于开发相关模型。模型是在 36 个分子的训练集上生成的。模型的相关系数 (r(2) ) 为 0.86 至 0.92，显著的交叉验证相关系数 (q(2) ) 为 0.79 至 0.85，F 检验值为 63.2 至 93.06，外部测试集的 r(2) 值 (pred_r(2) ) 为 0.69，预测数据集的相关系数 (pred r(2) Se) 为 0.77，自由度为 27 至 30。发现独立于对齐的描述符、SsOHE-index、SaaCHE index、SssCH2 和 x log P 是开发用于预测 HIV 整合酶抑制活性的相关模型的最重要描述符。
