Evolutionary-guided advanced deep-learning architecture powers mammalian GPCRome agonist predictions

Area of Science:

• Computational biology
• Pharmacology
• Genomics

Background:

• G-protein-coupled receptors (GPCRs) are crucial signaling proteins.
• Current computational methods for ligand-receptor interactions have limitations in receptor coverage, odorant receptor inclusion, and cross-species data.

Purpose of the Study:

• To present EvOlf, an evolutionary-guided deep-learning framework for predicting ligand-receptor interactions.
• To overcome limitations of existing methods by integrating diverse receptor types and extensive cross-species data.

Main Methods:

• Developed EvOlf, a deep-learning framework using hierarchical self-attention, multi-resolution feature encoding, and a fusion transformer.
• Utilized an expansive dataset of 105,235 experimentally validated interactions across 24 mammalian species.
• Integrated both odorant and non-odorant receptors.

Main Results:

• EvOlf achieved state-of-the-art performance in predicting ligand-receptor interactions.
• EvOlf significantly outperformed existing prediction methods on an independent test dataset.
• Experimental validation identified three novel allosteric co-activators for the β1-adrenergic receptor, enhancing agonist-evoked signaling.

Conclusions:

• EvOlf offers a robust and scalable platform for receptor deorphanization and novel ligand discovery.
• The framework's evolutionary approach and comprehensive dataset enhance prediction accuracy and biological relevance.
• Identified co-activators demonstrate EvOlf's potential for therapeutic target identification.