Abstract A total of 26 experimentally obtained retention indices on a logarithmic scale (log RI) from Thymus serpyllum essential oil were used to build a robust predictive model. The selected descriptors were used as inputs of an artificial neural network model to build a predictive quantitative structure-retention time relationship model. The coefficient of determination for the training cycle was 0.977, indicating that this model could be used for the prediction of retention indices for T. serpyllum essential oil compounds. These 26 compounds comprise about 99.8 % of the total oil, but among them only 6 compounds had the average relative concentration over 1 percent: geraniol (63.4 %), nerol (or cis-geraniol) (18.9 %), geranyl acetate (4.7 %), trans- caryophyllene (4.6 %), β-bisabolene (2.0 %) and geranial (1.2 %). According to these results, it can be concluded that T. serpyllum from village Sesalac (Serbia) belongs to geraniol chemotype, in total 82.3 % in both, trans and cis forms (63.4 % and 18.9 %, respectively).

中文翻译：

---

用于预测百里香精油香叶醇化学型保留指数的 QSRR 模型

摘要 使用来自百里香精油的总共 26 个实验获得的对数标度 (log RI) 保留指数来构建稳健的预测模型。选定的描述符用作人工神经网络模型的输入，以构建预测性定量结构保留时间关系模型。训练周期的决定系数为 0.977，表明该模型可用于预测 T. serpyllum 精油化合物的保留指数。这 26 种化合物约占总油的 99.8%，但其中只有 6 种化合物的平均相对浓度超过 1%：香叶醇 (63.4%)、橙花醇（或顺式香叶醇）（18.9%）、乙酸香叶酯（4.7%） )、反式石竹烯 (4.6 %)、β-红没药烯 (2.0 %) 和香叶醛 (1.2 %)。根据这些结果，