一种基于进化学习的中药定量组效关系建模方法

Approach to bioactivity prediction of hebal medicine using evolutional learning-based quantitative composition-activity relationship model

针对中药的组分配伍优化设计,提出一类中药药效智能预测及定量组效关系(QCAR)建模方法.采用全新设计的自适应浮点编码遗传算法,并与反向传播(BP)算法集成用于神经网络学习训练,既改善了神经元网络获得全局最优解的能力,又消除了经典遗传算法(GA)在编码与解码时产生的截断误差,从而提高了QCAR模型的训练与预报精度.以中药当归为实例对该方法进行应用研究.结果表明,无论是交叉验证还是独立检验均取得较好效果,明显优于常规建模方法,可用于对中药药效进行智能预测.

An evolutional learning method was proposed to establish a quantitative composition-activity relationship(QCAR)model and predict the bioactivity of Chinese medicine. Integrated with back propagation(BP)algorithm, a novel adaptive float coding genetic algoithm(AFCGA)was used to train neural network in QCAR modeling. The presented method improves the capability of obtaining the global optimum in neural computation and eliminates the truncation error produced in the process of encoding and decoding for classic genetic algorithms(GA). Hence, the training and predicting accuracy of the OCAR model was improved greatly. As an example,study of DangGui verified the predicting abililty of the proposed method. The results showed that, no matter in Cross Validation or Independent Test, the performance of the presented model significantly excels that of other conventional models. And this approach can be used to predict bioactivity of natural drugs from chemical analysis data.