高等学校化学学报 ›› 2025, Vol. 46 ›› Issue (3): 20240442.doi: 10.7503/cjcu20240442

• 物理化学 • 上一篇    下一篇

基于机器学习与分子动力学模拟发现CDK2抑制剂

谭英佳1, 陈亮2, 刘聿琳1, 那日松3, 赵熹1()   

  1. 1.吉林大学化学学院,理论化学研究所,长春 130023
    2.吉林省农业科学院大豆研究所,长春 130033
    3.河南农业大学植物保护学院,郑州 450046
  • 收稿日期:2024-09-25 出版日期:2025-03-10 发布日期:2024-11-19
  • 通讯作者: 赵熹 E-mail:zhaoxi@jlu.edu.cn
  • 基金资助:
    国家自然科学基金(32472613);河南省杰出青年基金项目(232300421008)

Discovery of CDK2 Inhibitors Based on Machine Learning and Molecular Dynamics Simulations

TAN Yingjia1, CHEN Liang2, LIU Yulin1, NA Risong3, ZHAO Xi1()   

  1. 1.College of Chemistry,Institute of Theoretical Chemistry,Jilin University,Changchun 130023,China
    2.Soybean Research Institute,Jilin Academy of Agricultural Sciences,Changchun 130033,China
    3.College of Plant Protection,Henan Agricultural University,Zhengzhou 450046,China
  • Received:2024-09-25 Online:2025-03-10 Published:2024-11-19
  • Contact: ZHAO Xi E-mail:zhaoxi@jlu.edu.cn
  • Supported by:
    the National Natural Science Foundation of China(32472613);the Excellent Youth Foundation of Henan province, China(232300421008)

摘要:

通过机器学习和分子动力学模拟方法发现了细胞周期蛋白依赖性激酶2(CDK2)潜在的抑制剂. 首先, 利用现有的大型活性数据库和机器学习算法, 建立了针对CDK2抑制剂的分类模型. 采用圆形指纹(ECFP6)的极端梯度提升树模型(XGBoost)筛选Enamine数据库, 并选出了1152个新型化合物. 通过分子对接和打分函数对这些潜在化合物在CDK2中的亲和力进行了排序, 并采用指纹聚类的方法将化合物分为4类. 分别从 4类中选择1种对接评分较高的化合物, 然后对4种化合物进行了类药性分析和分子动力学模拟. 类药性分析结果表明, 筛选出的4种潜在的CDK2抑制剂(Z1766368563, Z363564868, Z1891240670和Z2701273053)具有良好的成药性, 并在分子动力学模拟结果中具有较高的结合自由能. 这4种化合物可作为CDK2的先导化合物进行后续的改造和优化.

关键词: CDK2抑制剂, 机器学习, 分子动力学, 结合自由能

Abstract:

Four potential cyclin-dependent kinase 2(CDK2) inhibitors were discovered through machine learning and molecular dynamics simulation methods. First, a classification model for CDK2 inhibitors was established using existing large-scale activity databases and machine learning algorithms. The extreme gradient boosting(XGBoost) model with extended-connectivity fingerprints(ECFP6) was used to screen the Enamine database, identifying 1152 novel compounds. These potential compounds were then ranked based on their affinity for CDK2 using molecular docking and scoring functions. The compounds were clustered into four categories using fingerprint clustering methods, and one compound with a high docking score was selected from each category. Subsequently, the four selected compounds underwent drug-likeness analysis and molecular dynamics simulations. The four potential CDK2 inhibitors(Z1766368563, Z363564868, Z1891240670 and Z2701273053) demonstrated good drug-likeness properties and high binding free energy in molecular dynamics simulation results. The findings suggest that these four compounds can serve as lead compounds for subsequent modification and optimization as CDK2 inhibitors.

Key words: Cyclin-dependent kinase 2(CDK2) inhibitor, Machine learning, Molecular dynamics, Binding free energy

中图分类号: 

TrendMD: