Consensus QSAR modelling of SIRT1 activators using simplex representation of molecular structure

Hierarchical QSAR technology (HiT QSAR) was used for consensus QSAR modelling of 65 SIRT1 activators. Simplex representation of molecular structure (SiRMS) has been used for descriptor generation. The predictive QSAR models were developed using the partial least squares (PLS) method. The QSAR models were built up according to OECD principles. One hundred rounds of <i>Y</i>-scrambling were performed for each selected model to exclude chance correlations. A successful consensus model (<i>r</i>² = 0.830, = 0.754) was obtained from the five best QSAR models. Leverage, ellipsoid and local tree domain of applicability (DA) approaches have been used for evaluation of the quality of predictions. Molecular fragments responsible for an increase and decrease of the activation properties have been determined by mechanistic interpretation of the developed QSAR model.

本研究采用分层定量构效关系（Hierarchical QSAR, HiT QSAR）技术，针对65种SIRT1激活剂构建一致性定量构效关系模型。采用分子结构单纯形表示法（Simplex representation of molecular structure, SiRMS）进行分子描述符的生成。基于偏最小二乘（Partial Least Squares, PLS）方法构建预测型定量构效关系模型，所有模型的构建均遵循经济合作与发展组织（Organization for Economic Co-operation and Development, OECD）原则。针对每一个筛选获得的模型开展100轮Y随机化检验，以排除模型的偶然性相关。从5个最优定量构效关系模型中整合得到性能优良的一致性模型，其决定系数r²为0.830，另一评价系数为0.754。采用杠杆法、椭球法及局域树应用域（Domain of Applicability, DA）三种方法对预测结果的可靠性进行评估。通过对所构建的定量构效关系模型进行机理解析，明确了对SIRT1激活活性具有增强或减弱效应的分子片段。