OpioidBias: A Machine Learning Tool for Predicting the Biased Agonism of Opioid Ligands

Biased signaling at G-protein-coupled receptors (GPCRs)
enables
pathway-selective modulation but remains challenging to characterize
experimentally. We present OpioidBias, a machine learning tool for
predicting G-protein versus β-arrestin bias in opioid ligands.
A curated data set of opioid ligands was represented with >3800
descriptors
from RDKit and Mordred, encompassing physicochemical, topological,
and fingerprint-based features. Feature selection using Boruta and
recursive feature elimination (RFE) guided the training of six classifiers.
A random forest model incorporating combined RDKit and Mordred descriptors,
fingerprints, and RFE showed the best performance and was further
interpreted using feature analysis to identify molecular determinants
of bias. OpioidBias is freely available (http://github.com/PGlab-NIPER/OpioidBias) to support biased ligand discovery across opioid pharmacology.