Two data sets of natural antiviral agents including 107 anti-HIV1 and 18 anti-polio molecules were collected and subjected to quantitative structure-activity relationship (QSAR) analyses. A wide variety of molecular descriptors belonging to various structural properties were calculated for each molecule. Multiple linear regression (MLR) based on stepwise variable selection was employed to find the most convenient quantitative models. For each antiviral data set different QSAR models were established in two steps. Firstly, for each type of molecular descriptors separate QSAR analysis was performed, and then a new QSAR model was calculated using the selected descriptors in the first phase. For both types of antiviral data sets significant QSAR models were obtained. The atom-centered fragment descriptors represented the highest impact on the anti-HIV1 activity whereas for anti-polio agents, radial distribution function and three-dimensional MoRSE descriptors showed the most significant influences. Cross-validation and a separate prediction set were used to evaluate the stability and prediction ability of the models. It was found the discovered QSAR models for anti-HIV1 and anti-polio agents could reproduce about 80% and 90% of variances in the antiviral activity data with root mean square error of prediction of 0.421 and 0.171, respectively.

中文翻译：

---

QSAR研究天然来源的抗病毒化合物

收集了包括107种抗HIV1和18种抗脊髓灰质炎分子的两种天然抗病毒药数据，并进行了定量构效关系（QSAR）分析。为每个分子计算了属于各种结构特性的各种各样的分子描述符。使用基于逐步变量选择的多元线性回归（MLR）来找到最方便的定量模型。对于每个抗病毒数据集，分两个步骤建立了不同的QSAR模型。首先，对每种类型的分子描述符进行单独的QSAR分析，然后在第一阶段使用选定的描述符计算新的QSAR模型。对于两种类型的抗病毒数据集，均获得了重要的QSAR模型。以原子为中心的片段描述符对抗​​HIV1活性的影响最大，而对于抗策略药，径向分布函数和三维MoRSE描述符表现出最大的影响。使用交叉验证和单独的预测集来评估模型的稳定性和预测能力。发现发现的针对抗HIV1和抗策略药物的QSAR模型可以再现抗病毒活性数据中约80％和90％的方差，均方根误差分别为0.421和0.171。