Optimal Modeling of Anti-Breast Cancer Candidate Drugs Based on SSA-BP
Download as PDF
At present, in the research of candidate drugs of anti-breast cancer, the method of establishing quantitative structure activity relationship (QSAR) of compound activity is usually used to screen important compound molecular descriptors in order to save time and cost. Considering the nonlinear relationship between molecular descriptors and biological activities of compounds, neural network prediction models with strong nonlinear mapping ability and high accuracy have been widely used. However, there is no consensus on the best algorithm for QSAR modeling. In this paper, 11 molecular descriptors were firstly screened by nonlinear dimension reduction technique, and 11 molecular descriptors which have significant influence on the biological activity of compounds were selected. What’s more, QSAR models were constructed based on three traditional neural network models (BP neural network, Elman neural network and wavelet neural network) and neural network model improved by optimization algorithm (SSA-BP neural network). The results showed that the prediction error of SSA-BP neural network was the lowest, the mean square error was only 0.04898, and R2 reached 0.94572. Compared with the suboptimal BP neural network, MAE value decreased by 15.3%, MSE value decreased by 25.3%, MAPE value decreased by 37.4%, which indicates that BP neural network can predict the biological activity of compounds more accurately, and the optimization model can further improve the prediction performance of BP neural network. It is helpful to better screen efficient compound molecules and guide the structural optimization of existing active compounds and development of drugs for treating breast cancer.
Molecular descriptor, Dimension reduction, Qsar model, Bp neural network, Sparrow optimization algorithm