Exploring putative drug properties associated with TNF-alpha inhibition and identification of potential targets in cardiovascular disease using Machine Learning-Assisted QSAR Modeling and Virtual Reverse Pharmacology approach

Abstract

Introduction Cardiovascular disease is a chronic inflammatory disease with several categories of risk factors that impart a high mortality rate. Despite TNF-alpha being a prominent pro-inflammatory cytokine associated with chronic inflammation within cardiovascular disease, the adverse effects of current TNF-alpha based medications prompt an urgent need to identify efficient inhibitors as alternatives. This study not only explores the quantitative structural activity relationship (QSAR) of TNF-alpha inhibitors but also identifies potential drug targets to treat cardiovascular disease. Materials and Methods A GitHub Repository-based pipeline was used to curate data from the ChEMBL database. This was followed by pre-processing to exclude remove TNF-alpha inhibitors with missing bioactivity values and identify significant properties of molecules using exploratory data analysis (EDA). The extracted molecules were subjected to PubChem (PC) and SubStructure (SS) fingerprint descriptors, and a QSAR-based Random Forest model (QSAR-RF) was generated using the WEKA tool. QSAR-RF was validated using FDA drugs and molecules from PubChem and ZINC databases and used to predict the pIC50 value of the molecules selected from the docking study followed by molecular dynamic simulation with a time step of 100ns. Through virtual reverse pharmacology, we determined the main drug targets for the top four hit compounds obtained via molecular docking study. Our analysis included an integrated bioinformatics approach to pinpoint potential drug targets, as well as a PPI network to investigate critical targets. To further elucidate the findings, we utilized g:Profiler for GO and KEGG pathway analysis, ultimately identifying the most relevant cardiovascular disease-related pathway for the hub genes involved. Results A unique pipeline was used to create QSAR-RF a machine-learning model that identifies TNF-alpha inhibitors based on molecular features. It distinctly used PC and SS fingerprints, which show strong correlation coefficients of 0.993 and 0.992 respectively, with 0.607 and 0.716 as the respective 10-fold cross-validation scores. The VIP method extracts important features for each model. The QSAR-RF model was built using SS-fingerprints, and validated by docking study and small molecule bioactivity prediction. Irinotecan showed strong binding to TNF-alpha, with three important inhibitory features identified using a comprehensive variance importance plot (VIP). MD simulation confirmed the structural stability of the Irinotecan-TNF-alpha complex. For, the reverse network pharmacology approach, we identified four scaffolds namely, Tirilazad, Irinotecan, Diosgenin, and Gitogenin with higher binding scores. As a result, a total of 289 potential drug targets were identified for cardiovascular diseases (CVD). PPI network analysis identified EGRF, HSP900A1, STAT3, SRC, AKT1, MDM2, and other possible CVD targets. The treatment of CVD using four different scaffold drug targets was found to involve in oxidative stress, smooth muscle proliferation, organonitrogen compound, and multiple pathways such as PI3K-AKT signaling, lipid and atherosclerosis, among others. Conclusion In conclusion, Our study applies a ligand-based drug design approach to generate a SubStructure-based QSAR-RF prediction model to unravel the structural inhibitory feature of TNF-alpha inhibitors. And also identified multiple targets to treat CVD through a reverse network pharmacology approach.