基于机器学习的环氧合酶-2抑制剂分类模型的构建

onstruction of a Classification Model for Cyclooxygenase-2 Inhibitors based on Machine Learning

目的:构建环氧合酶-2(Cyclooxygenase-2,COX-2)抑制剂分类模型,用以筛选和优化COX-2抑制剂。方法:基于八种机器学习算法构建模型,比较不同模型的预测性能,筛选出最优模型后利用Y随机验证法对其进行测试,最后运用SHAP(Shapley Additive eXplanation)算法对最优模型进行可解释性分析。结果:八种不同模型的性能比较结果显示,基于随机森林算法建立的模型最优,其预测准确率、平衡准确率、马修斯相关系数、特征曲线下面积和F1分数(分别为0.893、0.825、0.673、0.909和0.933)最高;Y随机验证结果表明最优模型的预测结果并非偶然;此外,通过SHAP算法挖掘出20个最有可能影响COX-2抑制剂活性的结构片段。结论:本研究为新型COX-2抑制剂的开发提供理论依据,可供本领域其他研究人员对先导化合物进行优化或设计更好的COX-2抑制剂。

Objective:This study aims to develop a classification model for cyclooxygenase-2(COX-2)inhibitors for the purpose of screening and optimizing COX-2 inhibitors.Methods:Eight machine learning algorithms were used to construct models,and their predictive performance was compared to identify the best model.The optimal model was tested by using Y-scrambling validation method,finally the interpretability analysis of the optimal model was performed by using Shapley Additive eXplanation(SHAP)algorithm.Results:Among the eight different models compared,the Random Forest algorithm exhibited the best performance.With the highest accuracy,balanced accuracy,Matthew's correlation coefficient,area under the ROC curve,and F1 scores(0.893,0.825,0.673,0.909 and 0.933,respectively),it comes out on top.Validation with Y-scrambling showed that the predictions of the optimal model were not coincidence.Moreover,the SHAP algorithm was used to mine 20 structural fragments that could affect COX-2 inhibitor activity.Conclusion:In this study,we developed a theoretical basis for developing COX-2 inhibitors,which is useful to other researchers in this field when optimizing lead compounds and designing new COX-2 inhibitors.