Docking and AI-based classification cum recurrent neural network validation for predicting anti-tubercular leads against dprE-1 enzyme targeter

Tuberculosis (TB) continues to pose a major global health challenge, highlighting the need for the development of new therapeutic agents. The lengthy and complex course of treatment for tuberculosis involves several medications and frequently leads to non-compliance and severe adverse effects. This research employs combined in silico and Artificial Intelligence-Machine Learning (AI-ML) strategy to identify potential anti-TB compounds. Initial compound screening was performed using tools like SwissADME for pharmacokinetic and drug-likeness analysis, Molinspiration for bioactivity profiling, PASSWAY for biological activity prediction and Protox-3.0 to estimate LD50 values ensuring compound safety. Molecular docking studies were conducted using SwissDock to evaluate ligand-receptor interactions, with the best compounds selected based on the most negative binding free energy (ΔG) values. For visualization UCSF Chimera & Biovia Discovery were used. To further validate the findings, artificial intelligence and machine learning (AI-ML) techniques, including Support Vector Machines (SVM), Gaussian Naïve Bayes, k-nearest Neighbors (KNN), were applied for predictive analysis of compound efficacy. Asp-pro-lys, Thr-ser-pro, Glu-trp, Phe-ala, Cys-lys, Gln-phe were the best leads having docking score ΔG [kcal/mol] -9.9492, -9.9451, -9.3372,-9.2957,-9.1472 and -9.0621 respectively. According to in-silico and AI-ML studies there was 63.64% accuracy in SVM & k-nearest Neighbors while Gaussian Naïve Bayes had 59.09% accuracy. Recurrent Neural Network (RNN) was used to validate the LD50 values which had 12.5% accuracy. The integrated approach highlights the potential of combining in silico methodologies with AI-ML for efficient and reliable anti-TB drug discovery, paving the way for future experimental validations and clinical applications.