College of Chemistry & Chemical Engineering, Lanzhou University, Lanzhou, China.
Chem Biol Drug Des. 2020 Feb;95(2):240-247. doi: 10.1111/cbdd.13637. Epub 2019 Nov 13.
Non-structural viral protein 5B (NS5B) is a viral protein in hepatitis C virus. Although various inhibitors against NS5B have been found, the activity prediction of similar untested inhibitors is still highly desirable. In this respect, the Tchebichef moments (TMs) calculated from the images of molecular structures were regarded as the independent variables while the inhibitory activity (pIC ) was the dependent variable, and the predictive model was established by means of stepwise regression. The R-squared of leave-one-out cross-validation (Q ) for the training set and the R-squared of prediction ( ) for external independent test set were 0.919 and 0.927, respectively. The obtained model was also evaluated strictly. Compared with the multivariate curve resolution with alternating least squares (MCR-ALS) and the QSAR approaches derived from the literature, the proposed method is more accurate and reliable. This study not only provides an effective approach to predict the biological activity of RNA replication's inhibitors, but also extends the QSAR modeling technique.
非结构病毒蛋白 5B（NS5B）是丙型肝炎病毒中的一种病毒蛋白。尽管已经发现了各种针对 NS5B 的抑制剂，但仍然非常需要对类似未经测试的抑制剂的活性进行预测。在这方面，从分子结构图像计算出的 Chebyshev 矩（TM）被视为自变量，而抑制活性（pIC）则为因变量，并通过逐步回归建立预测模型。训练集的留一交叉验证的 R 平方（Q）和外部独立测试集的预测 R 平方（）分别为 0.919 和 0.927。所得到的模型也经过了严格的评估。与交替最小二乘法的多元曲线分辨（MCR-ALS）和来自文献的 QSAR 方法相比，所提出的方法更加准确可靠。本研究不仅提供了一种有效的方法来预测 RNA 复制抑制剂的生物活性，还扩展了 QSAR 建模技术。
