|
|
|
| Virtual screening model for fibroblast growth factor receptors kinase inhibitors based on machine learning |
| DING Juntao 1, LIU Bo 1, WU Jianzhang 2, LI Wulan 1. |
1.The First School of Medicine, School of Informationand Engineering, Wenzhou Medical University, Wenzhou 325035, China; 2.The Eye Hospital of Wenzhou Medical
University, Wenzhou 325000, China |
|
| Cite this article: |
|
DING Juntao,LIU Bo,WU Jianzhang, et al. Virtual screening model for fibroblast growth factor receptors kinase inhibitors based on machine learning[J]. JOURNAL OF WEZHOU MEDICAL UNIVERSITY, 2023, 53(7): 548-555，564.
|
|
|
Abstract Objective: To construct an efficient virtual screening model to filter fibroblast growth factor receptors (FGFR) kinase inhibitors based on machine learning. Methods: FGFR kinase inhibitors from the public dataset BindingDB were collected; RDkit was used to calculate molecular descriptors to characterize compound molecules for data input. Two machine learning algorithms (random forest and support vector machine) were used to establish a virtual screening model, and four indicators [accuracy, precision, recall and area under curve (AUC)] were used to evaluate the model. Preliminary screening of 13 million compounds using a random forest
model; Subsequently, Autodock Vina and Glide methods were used to further screen FGFR1 kinase inhibitors; Molecular dynamics simulation was used to analyze the compounds obtained through virtual screening. Results: The constructed random forest model and support vector machines model had good performance in terms of accuracy, precision, recall and AUC. The accuracy and AUC of the random forest model reached 0.878 and 0.952 respectively, which could be used as a virtual screening model for FGFR kinase inhibitors. When the random forest model was used in the high-throughput virtual screening to obtain highly active lead compound, the molecular docking and molecular dynamics analysis of the three selected optimal compounds with FGFR1 kinase showed that there was a high similarity with the positive drug AZD4547 in terms of hydrogen bonding, binding free energy,and hydrophobic effect. The LEU21, VAL29 and ALA49 residues of FGFR1 are important residues for small molecule drugs to maintain stable binding. Conclusion: This study provides a virtual screening model for FGFR kinase inhibitors based on machine learning, which can be used for efficient screening of large-scale small
molecule compound libraries.
|
|
|
| |
|
|
|
|
|
|
