INTEGRATIVE COMPUTATIONAL FRAMEWORK FOR DOPAMINE D3 RECEPTOR-TARGETED DRUG DISCOVERY: BRIDGING GENETIC VARIABILITY, STRUCTURAL DYNAMICS, AND MACHINE LEARNING

Abstract

Aim: The Dopamine D3 receptor (D3R), a G-protein coupled receptor predominantly expressed in the limbic system, plays a pivotal role in modulating reward, cognition, and emotional behaviors, and is implicated in neuropsychiatric and neurodegenerative disorders such as Parkinson’s disease and schizophrenia. Given its therapeutic relevance, this study presents an integrated computational framework to elucidate the structural and functional consequences of D3R mutations, their impact on ligand binding, and the application of advanced machine learning for drug discovery. Evolutionary conservation analysis using ConSurf identified functionally critical regions within D3R, while PredictSNP predicted deleterious effects for 73 out of 405 single amino acid variants, with a majority located in highly conserved regions. To further probe the dynamic consequences of mutation, molecular dynamics (MD) simulations were performed for the Wild-type and selected variants (P344T, L60P) over 100 ns. A cascade neural network-based quantitative structure-activity relationship (QSAR) model was developed and trained on a large, chemically diverse dataset. The model utilized a two-stage architecture with uncertainty estimation and active learning, leveraging molecular fingerprints and 2D/3D descriptors. Performance evaluation demonstrated robust classification of high- and low-affinity ligands, with high ROC-AUC (0.888), average precision (0.916), and strong rank correlation (Spearman 0.780). The model’s precision-recall and ROC curves indicated high discriminative power, while confusion matrix analysis revealed a conservative bias, minimizing false positives at the cost of some false negatives-a trade-off often desirable in early-stage drug discovery. Results and Conclusion: The study identified 73 deleterious D3R variants, predominantly in conserved regions, correlating with reduced dopamine binding affinity and altered receptor dynamics. Molecular docking and MD simulations confirmed that mutations in critical regions impair function, while those in variable regions are generally tolerated. The cascade neural network QSAR model achieved high accuracy (79.7%), precision (91.6%), and ROC-AUC (0.888), effectively distinguishing high- and low-affinity ligands. Misclassifications were mainly confined to borderline cases, indicating robust model calibration. This integrative computational approach provides a powerful platform for D3R-targeted drug discovery, supporting the rational design and prioritization of novel therapeutics for neuropsychiatric and neurodegenerative disorders, with future work focusing on model interpretability and experimental validation.