Multi-target directed ligand-based 2D-QSAR models were developed using different N-benzyl piperidine derivatives showing inhibitory activity toward acetylcholinesterase (AChE) and β-Site amyloid precursor protein cleaving enzyme (BACE1). Five different classes of molecular descriptors belonging to spatial, structural, thermodynamics, electro-topological and E-state indices were used for machine learning by linear method, genetic function approximation (GFA) and nonlinear method, support vector machine (SVM) and artificial neural network (ANN). Dataset used for QSAR model development includes 57 AChE and 53 BACE1 inhibitors. Statistically significant models were developed for AChE (R² = 0.8688, q² = 0.8600) and BACE1 (R² = 0.8177, q² = 0.7888) enzyme inhibitors. Each model was generated with an optimum five significant molecular descriptors such as electro-topological (ES_Count_aaCH and ES_Count_dssC), structural (QED_HBD, Num_TerminalRotomers), spatial (JURS_FNSA_1) for AChE and structural (Cl_Count, Num_Terminal Rotomers), electro-topological (ES_Count_dO), electronic (Dipole_Z) and spatial (Shadow_nu) for BACE1 enzyme, determining the key role in its enzyme inhibitory activity. The predictive ability of the generated machine learning models was validated using the leave-one-out, Fischer (F) statistics and predictions based on the test set of 11 AChE (r² = 0.8469, r²_pred = 0.8138) and BACE1 (r² = 0.7805, r²_pred = 0.7128) inhibitors. Further, nonlinear machine learning methods such as ANN and SVM predicted better than the linear method GFA. These molecular descriptors are very important in describing the inhibitory activity of AChE and BACE1 enzymes and should be used further for the rational design of multi-targeted anti-Alzheimer’s lead molecules.

Graphic abstract

中文翻译：

---

预测 AChE 和 BACE1 双抑制剂治疗阿尔茨海默病活性的机器学习模型

使用显示对乙酰胆碱酯酶 (AChE) 和β-位点淀粉样前体蛋白裂解酶 (BACE1)的抑制活性的不同 N-苄基哌啶衍生物开发了基于多靶点定向配体的 2D-QSAR 模型。属于空间、结构、热力学、电拓扑和E状态指数的五种不同类别的分子描述符通过线性方法、遗传函数逼近 (GFA) 和非线性方法、支持向量机 (SVM) 和人工神经网络用于机器学习网络（ANN）。用于 QSAR 模型开发的数据集包括 57 种 AChE 和 53 种 BACE1 抑制剂。为 AChE 开发了具有统计学意义的模型 ( R ²  = 0.8688, q ² = 0.8600) 和 BACE1 ( R ²  = 0.8177, q ²  = 0.7888) 酶抑制剂。每个模型都使用最佳的五个重要分子描述符生成，例如电拓扑（ES_Count_aaCH 和 ES_Count_dssC）、结构（QED_HBD、Num_TerminalRotomers）、AChE 的空间（JURS_FNSA_1）和结构（Cl_Count、Num_Terminal Rotomers）、电拓扑（ES_Count_dO） 、电子（Dipole_Z）和空间（Shadow_nu）为BACE1酶，决定了其酶抑制活性的关键作用。使用留一法、Fischer ( F ) 统计和基于 11 AChE 测试集的预测验证了生成的机器学习模型的预测能力 ( r ²  = 0.8469, r² _pred  = 0.8138) 和 BACE1 ( r ²  = 0.7805, r ² _pred  = 0.7128) 抑制剂。此外，ANN 和 SVM 等非线性机器学习方法的预测效果优于线性方法 GFA。这些分子描述符对于描述 AChE 和 BACE1 酶的抑制活性非常重要，应进一步用于多靶点抗阿尔茨海默氏症先导分子的合理设计。

图形摘要