Li Yali, Wang Ling, Liu Zhihong, Li Chanjuan, Xu Jiake, Gu Qiong, Xu Jun
Research Center for Drug Discovery & Institute of Human Virology, School of Pharmaceutical Sciences, Sun Yat-Sen University, Guangzhou 510006, China.
Mol Biosyst. 2015 May;11(5):1241-50. doi: 10.1039/c4mb00718b.
Liver X receptor (LXR) α and β are cholesterol sensors; they respond to excess cholesterol and stimulate reverse cholesterol transport. Activating LXRs represents a promising therapeutic option for dyslipidemia. However, activating LXRα may cause unwanted lipogenicity. A better anti-dyslipidemia strategy would be to develop selective LXRβ agonists that do not activate LXRα. In this paper, a data set of 234 selective and non-selective LXRβ agonists was collected from the literature. For the first time, we derived the classification models from the data set to predict selective LXRβ agonists using multiple machine learning methods (naïve Bayesian (NB), Recursive Partitioning (RP), Support Vector Machine (SVM), and k-Nearest Neighbors (kNN) methods) with optimized property descriptors and structural fingerprints. The models were optimized from 324 multiple machine learning models, and most of the models showed high predictive abilities (overall predictive accuracies of >80%) for both training and test sets. The top 15 models were evaluated using an external test set of 76 compounds (all containing new scaffolds), and 10 of them displayed overall predictive accuracies exceeding 90%. The top models can be used for the virtual screening of selective LXRβ agonists. The NB models can identify privileged and unprivileged fragments for selective LXRβ agonists, and the fragments can be used to guide the design of new selective LXRβ agonists.
肝脏X受体（LXR）α和β是胆固醇传感器；它们对过量胆固醇作出反应并刺激胆固醇逆向转运。激活LXR是治疗血脂异常的一种有前景的治疗选择。然而，激活LXRα可能会导致不必要的脂肪生成。一种更好的抗血脂异常策略是开发不激活LXRα的选择性LXRβ激动剂。在本文中，从文献中收集了一个包含234种选择性和非选择性LXRβ激动剂的数据集。我们首次使用多种机器学习方法（朴素贝叶斯（NB）、递归划分（RP）、支持向量机（SVM）和k近邻（kNN）方法）以及优化的性质描述符和结构指纹，从该数据集中推导分类模型以预测选择性LXRβ激动剂。这些模型是从324个多种机器学习模型中优化得到的，并且大多数模型对训练集和测试集都显示出较高的预测能力（总体预测准确率>80%）。使用一个包含76种化合物（均含有新骨架）的外部测试集对前15个模型进行了评估，其中10个模型的总体预测准确率超过了90%。顶级模型可用于选择性LXRβ激动剂的虚拟筛选。NB模型可以识别选择性LXRβ激动剂的特权和非特权片段，这些片段可用于指导新型选择性LXRβ激动剂的设计。
