抗乳腺癌候选药物的ADMET性质的预测研究

Predictive Study of the ADMET Properties of Anti-Breast Cancer Drug Candidates

摘要: 近年来，乳腺癌已经成为全世界范围内女性患病率和死亡率非常高的恶性肿瘤，研究与制作抗乳腺癌药物已经迫在眉睫。在此背景下，本文主要研究了能够拮抗ERα活性的抗乳腺癌候选药物的ADMET (吸收Absorption、分布Distribution、代谢Metabolism、排泄Excretion和毒性Toxicity)性质的预测模型，对临床试验得到的1974个化合物的ADMET数据进行预处理和相关分析。运用BP神经网络和XGBoost回归两种方法建立并研究了两种对化合物ADMET性质的定量预测模型。实验研究结果表示，对比于BP神经网络方法，XGBoost分类预测模型对于该任务误差最低、效果最好。

Abstract: In recent years, breast cancer has become a malignancy with a very high prevalence and mortality rate in women worldwide, and the research and production of anti-breast cancer drugs has become urgent. In this context, this paper focuses on the prediction model of ADMET (Absorption Absorp-tion, Distribution Distribution, Metabolism Metabolism, Excretion Excretion and Toxicity) proper-ties of anti-breast cancer drug candidates capable of antagonizing ERα activity, for 1974 compounds obtained from clinical trials. The ADMET data were preprocessed and correlated. Two quantitative prediction models for the ADMET properties of the compounds were developed and investigated using both BP neural network and XGBoost regression methods. The results of the experimental study indicated that the XGBoost classification prediction model had the lowest error and the best results for this task compared to the BP neural network approach.