GPCR-A17 MAAP: mapping modulators, agonists, and antagonists to predict the next bioactive target

G Protein-Coupled Receptors (GPCRs) are vital players in cellular signalling and key targets for drug discovery, especially within the GPCR-A17 subfamily, which is linked to various diseases. To address the growing need for effective treatments, the GPCR-A17 Modulator, Agonist, Antagonist Predictor (MAAP) was introduced as an advanced ensemble machine learning model that combines XGBoost, Random Forest, and LightGBM to predict the functional roles of agonists, antagonists, and modulators in GPCR-A17 interactions. The model was trained on a dataset of over 3,000 ligands (agonists, antagonists, and modulators) and 6,900 protein-ligand interactions, comprising all three ligand types, sourced from the Guide to Pharmacology, Therapeutic Target Database, and ChEMBL. It demonstrated a strong predictive performance, achieving F1 scores of 0.9179 and 0.7151, AUCs of 0.9766 and 0.8591, and specificities of 0.9703 and 0.8789, respectively, reflecting the overall performance across all classes in the testing and independent ligand validation datasets. A Ki-filtered subset of 4,274 interactions (where Ki is the inhibition constant that quantifies the ligand-binding affinity) improved the F1 scores to 0.9330 and 0.8267 for the testing and independent ligand datasets, respectively. By guiding experimental validation, GPCR-A17 MAAP accelerates drug discovery for various therapeutic targets. The code and data are available on GitHub ( https://github.com/MoreiraLAB/GPCR-A17-MAAP ).