Intelligent consensus-based QSAR modeling and virtual screening of dual inhibitors of 5HT_1A/5HT₇ serotonin receptors

Dual inhibitors of 5HT_1A and 5HT₇ serotonin receptors provide improved therapeutic effectiveness for depression, anxiety disorders, and neuropsychiatric conditions, potentially decreasing the necessity for multiple medications and reducing side effects. This study presents a cutting-edge computational framework for the discovery and optimization of dual 5HT_1A/5HT₇ inhibitors, integrating machine learning (ML), molecular docking, and virtual screening. A meticulously curated dataset of dual inhibitors was employed to develop regression and classification models using state-of-the-art ML algorithms. The consensus regression model achieved remarkable predictive performance (R²_Test>0.93), and reduced RMSE_CV by 30-40 % compared to individual models. In classification tasks, the majority voting method boosts accuracy to 92 % and achieves a 25 % increase in F1 scores, surpassing individual models and showcasing strong generalization across 5-fold cross-validation and y-randomization tests. The validated consensus-based QSAR model was employed as endpoint analysis of virtual screening of dual 5HT_1A/5HT₇ inhibitors. This research highlights the power of consensus modeling and ensemble approaches in enhancing predictive accuracy and robustness. By integrating machine learning, structural insights, virtual screening and ADMET analysis, we provide a reliable and efficient strategy for identifying novel 5HT_1A/5HT₇ inhibitors, paving the way for future experimental validation and therapeutic development of next-generation treatments for serotonin-related disorders.