A Computational Approach to Identify Potent Inhibitors of Janus Kinase ۱ from Natural Products: Structure-Based High-Throughput Virtual Screening and LightGBM Classifier

Computational approaches play an important role in drug development because they reduce time and costs while boosting the likelihood of generating effective therapeutic candidates. The combination of several approaches, such as virtual screening and quantitative structure-activity relationship (QSAR) models, allows for the investigation of the link between molecular structure and biological activity against specific targets. In recent years, advances in computational tools have revolutionized the drug discovery environment, allowing researchers to effectively navigate enormous chemical landscapes and uncover potential candidates for clinical development. This study aims to find natural compounds from the IMPPAT database that effectively inhibit the JAK۱ protein. JAK۱ (Janus kinase ۱) is a pivotal enzyme in the JAK-STAT signaling pathway, influencing immune responses and hematopoiesis. Its dysregulation is associated with multiple diseases, such as atopic dermatitis, rheumatoid arthritis, and certain cancers. The critical role of JAK۱ in mediating signals from various cytokines underscores the importance of developing selective JAK۱ inhibitors for therapeutic intervention. By targeting JAK۱, it may be possible to modulate inflammatory responses and improve treatment outcomes for patients suffering from these conditions. This study's methodology includes several important phases to ensure a complete evaluation of potential JAK۱ inhibitors. Initially, the ۳D crystallographic structure of the JAK۱ protein was acquired from the PDB database. The structures of known JAK۱ inhibitors were collected from the PubChem database, and their structural fingerprints were computed using the RDkit tool in Python. In the next step, a LightGBM (LGBM) classifier was constructed and trained utilizing structure fingerprints to accurately assess the link between chemical structure and biological activity. Next, natural product structures were downloaded from the IMPPAT database, which offers a plethora of information on bioactive substances produced from natural sources, and Morgan fingerprints were calculated for them. High-throughput virtual screening (HTVS) has emerged as pivotal to screen thousands of chemical structures rapidly and efficiently is essential for identifying viable leads. The first step in our HTVS approach was employing molecular docking simulation to evaluate the binding affinities and quickly identifying compounds that exhibit favorable interactions with the JAK۱ protein, the next step was to apply Lipinski's Rule of Five as a filtering mechanism. This rule serves as a guideline for assessing the drug-likeness of compounds based on their molecular properties. The filtered compounds were then subjected to predictive modeling using machine learning classifiers. The best-performing structures were then selected for predictive modeling to assess their potential biological activity. Finally, the activity of selected structure against JAK۱ were predicted by LGBM classifier and four natural products (IMPHY۰۰۴۸۵۷, IMPHY۰۰۹۳۸۰, IMPHY۰۰۰۴۵۱, and IMPHY۰۰۸۸۷۴) were identified as promising inhibitors. These compounds demonstrate significant potential as effective JAK۱ inhibitors based on their predicted interactions and favorable pharmacokinetic properties. The study effectively identified four natural compounds as potential JAK۱ inhibitors, highlighting the significant role of computational approaches in drug discovery. However, it is essential to conduct experimental validation to confirm the biological activity and therapeutic efficacy of these identified structures.

کلیدواژه ها:

computational drug discovery ، virtual screening ، JAK۱ inhibitors