همایش ملی بیوانفورماتیک ایران

صفحه اصلی / 4th international edition and 13th Iranian Conference on Bioinformatics

Leveraging Machine Learning Models for Virtual Screening of ZINC Database to Identify JAK1 Inhibitors

نویسندگان :

Negar Abdolmaleki¹ Hamid Mahdiuni²

1- دانشگاه رازی 2- دانشگاه رازی

کلمات کلیدی :

Janus Kinase 1 Inhibitors،Virtual Screening،Machine Learning،MACCS Fingerprint،Random Forest

چکیده :

JAnus Kinase 1 (JAK1) inhibitors have emerged as promising therapeutic agents, playing crucial roles within the Janus kinase-signal transducer and activator of transcription (JAK-STAT) pathway. This pathway is vital for modulating immune responses and regulating inflammation, particularly in autoimmune diseases, inflammatory disorders, and cancers (Lv and Qi, 2024). The current investigation employed various machine learning models such as Support Vector Machine (SVM), XGBoost, and Random Forest (RF) to identify potential JAK1 inhibitors. The model training and validation dataset consisted of 7,374 JAK1 inhibitors, with SMILES (Simplified Molecular Input Line Entry System) strings sourced from ChEMBL (Gaulton and Hersey, 2017), BindingDB (Gilson and Liu, 2016), and PubChem (Kim and Chen, 2019). Each inhibitor was annotated with an activity label 1 for active JAK1 inhibition and 0 for inactive compounds. Input for the machine learning models was generated by processing reference compounds with RDKit, a cheminformatics toolbox, to extract molecular descriptors characterized by MACCS (Molecular ACCess System) fingerprinting (Kong and Huang, 2023). The models’ performance was evaluated using several metrics, including precision, F1 score, recall, specificity, and area under the curve (AUC). The achieved AUC-ROC scores were 0.97 for SVM, 0.981 for XGBoost, and 0.986 for the Random Forest model, demonstrating superior performance across all evaluation criteria. Due to its accuracy and robustness, the RF model was selected for virtual screening of the ZINC database (Sterling and Irwin, 2015.), a comprehensive repository of commercially available compounds. According to the predicted activity scores, the screening process identified several promising candidates as effective JAK1 inhibitors. Our findings emphasize the effectiveness of Random Forest combined with MACCS fingerprinting as a powerful method for the virtual screening of the kinase inhibitors. This approach facilitates the identification of potential JAK1 inhibitors for further experimental validation, potentially accelerating the discovery of new therapeutic agents for immunological and inflammatory diseases while providing insights for developing targeted therapies.