Quantitative Structure-Activity Relationship Modeling of Estrogen Receptor Alpha Bioactivity based on Multiple Algorithms

Currently, bioactivity analysis and ADMET property analysis of compound molecules in drug discovery are time-consuming and labor-intensive, so the approach of building compound bioactivity prediction models is usually used to screen potentially active compounds, especially the rapid development of data mining and machine learning methods has greatly facilitated this process. The aim of this paper is to construct quantitative prediction models for the bioactivity of compounds ERα. Firstly, 12 molecular descriptors that are strongly correlated with bioactivity and independent of each other were screened from 729 molecular descriptors to construct a quantitative prediction model of compound ERα bioactivity based on BP neural network, decision tree and random forest, respectively. Finally, the three models were evaluated with four evaluation metrics - variance, standard deviation, mean absolute error, and root mean square error. According to the comparison results, the best prediction model is the random forest model, which can be used to predict new compound molecules with better ERα bioactivity in the future.