Abstract: Screening for anti-breast cancer drug candidates is of great significance in the treatment of breast cancer. Antihormonal therapy for breast cancer is often used in breast cancer patients with ERα expression, and the higher the anti-ERα activity value, the more effective the drug is for treatment. In this paper, 729 molecular descriptors of the compound are filtered firstly using the gradient boost model XGBoost and the distance correlation coefficient matrix, then based on the filtered 20 molecular descriptors with their activity values and the genetic algorithm, a GA-BP neural network model is established, which has a mean squared error MSE=0. 105 and a coefficient of determination $\mathrm{R}^2=0. 946$, and therefore is a high-precision model for screening potential drugs based on data mining techniques. 

Key words: Screening of anti-breast cancer drug, Distance correlation coefficient, XGBoost algorithm, GA-BP neural network, Mean squared error