Explainable QSAR models of 5-HT1A receptor ligands using conceptual DFT descriptors and no-code machine learning...

Quantitative structure-activity relationship (QSAR) modeling is a central component of molecular informatics and ligand-based drug discovery, enabling the analysis and interpretation of structure-activity relationships from molecular descriptors. In this study, mechanistically interpretable binary QSAR models were developed to classify the activity of 89 serotonin 5-HT1A receptor ligands by integrating Conceptual Density Functional Theory (CDFT) descriptors with no-code machine learning workflows. Neutral and protonated molecular forms were considered under both vacuum and aqueous conditions, and electronic reactivity indices (η, μ, ω, and ΔN_max) were computed at the GFN1-xTB level of theory. Model construction employed the RandomTree algorithm within an OECD-consistent framework, using Ki values as biological endpoints. Across all decision trees and validation splits, the electrophilicity index of protonated molecules in aqueous solvent repeatedly emerged as the root node, suggesting a potential contribution of electrostatic complementarity under physiologically relevant conditions. Chemical hardness of neutral molecules in aqueous phase was identified as a secondary descriptor associated with activity classification, reflecting stability-related effects. External validation yielded a Cohen's Kappa value of 0.49, with precision, recall, and F1-score above 0.80, and AUC-ROC values between 0.69 and 0.73, indicating moderate but consistent exploratory classification performance within the studied applicability domain. These results suggest that explicitly accounting for protonation and solvation may improve the mechanistic interpretability and exploratory classification behavior of CDFT-based QSAR models for protonatable 5-HT1A ligands. The proposed no-code workflow provides a transparent and reproducible exploratory molecular informatics framework for exploratory and mechanistically interpretable ligand-based modeling in early-stage drug discovery.