Activity Prediction of Strong Quinazoline-Derived Drugs Using Machine Learning Methods

Cancer is the uncontrolled growth and proliferation of cells. Inhibition of protein kinase is vital in cancer treatment. Therefore, protein kinase inhibitors can act as drugs. In order to reduce the cost and enhance the speed in the design of drug molecules, the quantitative structure-activity relationship (QSAR) method was introduced. In this research, the QSAR was used to predict the half-maximal inhibitory concentration (IC۵۰) of ۱۲۹ molecules derived from quinazoline compounds to inhibit B-raf kinase. Molecules were divided into strong (IC۵۰ between ۲.۴ and ۳۱۷.۴) and weak (IC۵۰ between ۳۴۴.۸ and ۱۱۵۴) drugs using a normal probability plot. A model with ۹ descriptors based on genetic algorithm-multivariate linear regression (GA-MLR) has been presented for the strong drug category. In addition to this multivariable linear model, the QSAR was performed for the nonlinear model of feed-forward neural network (FFNN), support vector machine-particle swarm optimization (PSO-SVM), and decision tree (DT). For the strong drug, the coefficient of determination follows the trend of R۲ FFNN >R۲ PSO-SVM≈ R۲ DT> R۲ GA-MLR. As a result, the FFNN model shows the best performance with R۲=۹۴.۹۶ for the strong drug category compared to other models. The neural network has made a significant improvement compared to the linear model for the strong drug category. For a strong drug, the presence or absence of two carbon and nitrogen atoms in the topological distance of ۴ is the most effective molecular descriptor. Finally, the QSAR model presented in this research, in addition to the quantitative prediction of IC۵۰ to inhibit B-raf kinase, by identifying effective molecular descriptors, can help in the process of drug design.

کلیدواژه ها:

Quantitative structure-activity relationship (QSAR) ، Quinazoline derivative drugs ، B-raf kinase ، Half-maximal inhibitory concentration (IC۵۰) ، Genetic algorithm-multiple linear regression (GA-MLR) ، Machine learning