Revealing key structural features for developing new agonists targeting δ opioid receptor: Combined machine learning and molecular modeling perspective

Abstract

Despite being the most widely prescribed and misused type of medication, opioids continue to function as robust pain relief agents; however, overdosing is a significant cause of fatalities among opioid users. The δ-opioid receptor (DOR) has immense promise in treating long-term pain by producing anxiolytic and antidepressant-like outcomes. Although DOR agonists play a crucial role, their clinical implementation is restricted because of the probable manifestation of severe, life-threatening complications. A Python-based machine learning approach was employed to develop a quantitative structure–activity relationship (QSAR) model in this study. To address this, 4217 compounds and their associated biological inhibition activities were retrieved from the gpcrdb database. The K-best features selection method revealed three key structural features such as SLOGPVSA2, Chi6ch, and S17 contributed significantly to the best model performance. Statistical analysis, K-fold cross-validation, applicability domain analysis, and external validation using 38 unseen FDA-approved drug data confirmed the robustness of the predictive model. A molecular docking study in along with Ligand–Receptor Contact Fingerprints (LRCFs) using the essential chemical interactions described for analog ligands releaved the key contact interactions of Asp 128, Tyr 129, Met 132, Trp 274, Ile 277, and Tyr 308 residues in the total binding affinities upon complexation. Our combinatorial study using regression QSAR and ligand–receptor Contact, analysis could serve in the design of more rational compounds for drug discovery targeting DOR.

Keywords

• δ Opioid Receptor
• Machine learning
• QSAR
• Ligand–Receptor Contacts
• Non-Linear Regression
• XGBOOST