Predictive Models to Identify Small Molecule Activators and Inhibitors of Opioid Receptors

Area of Science:

• Computational chemistry and pharmacology
• Drug discovery and development
• Machine learning in medicinal chemistry

Background:

• Opioid receptors (OPRs) are key targets for pain management.
• Current opioid analgesics cause adverse effects and addiction.
• Novel, safer analgesics are urgently needed to address the opioid crisis.

Purpose of the Study:

• To develop predictive computational models for OPR activity based on chemical structures.
• To apply these models to screen large compound libraries for novel OPR-active molecules.
• To experimentally validate computationally identified compounds.

Main Methods:

• Utilized quantitative high-throughput screening (qHTS) data for three OPRs (agonist and antagonist modes).
• Developed and compared four machine learning algorithms, with Random Forest showing superior performance.
• Applied the best models to virtually screen compound libraries and experimentally validated predictions.

Main Results:

• The Random Forest model for mu OPR (OPRM) agonists demonstrated high performance (AUC-ROC 0.88, MCC 0.7).
• Virtual screening enriched hit rates by at least twofold compared to original assays.
• Identified and experimentally confirmed several novel potent OPR activators and inhibitors.

Conclusions:

• Developed robust OPR prediction models applicable for prioritizing compounds in large libraries.
• Successfully identified novel potent OPR ligands through computational screening and experimental validation.
• Molecular docking provided insights into the interactions of novel ligands with OPR active sites.