Temporally Resolved and Interpretable Machine Learning Model of GPCR conformational transition

Area of Science:

• Pharmacology and Computational Biology
• Neuroscience
• Drug Discovery

Background:

• Developing subtype-specific drugs for highly homologous proteins like dopamine receptors D2R and D3R presents a significant challenge in pharmacology.
• Distinct conformational ensembles in protein dynamics lead to differences in target-specific cryptic druggable sites, complicating drug design.

Purpose of the Study:

• To develop a scalable and unbiased data analysis strategy for pinpointing residue communities that regulate conformational state ensembles in protein dynamics.
• To identify distinct residue communities specific to Dopamine D3 receptor (D3R) conformational transitions compared to Dopamine D2 receptor (D2R).

Main Methods:

• Utilized Molecular Dynamics (MD) simulations to analyze protein dynamics.
• Developed and applied the Dynamically Resolved Universal Model for BayEsiAn network Tracking (DRUMBEAT), an interpretable machine learning algorithm.
• Validated DRUMBEAT by identifying residue communities involved in the deactivation of the β2-adrenergic receptor.

Main Results:

• Identified distinct and non-conserved residue communities specific to D3R conformational transitions compared to D2R.
• These key residue communities are located around the F1704.62–F172ECL2 and S1464.38–G14134.56 contacts.
• The DRUMBEAT algorithm was successfully validated on the β2-adrenergic receptor.

Conclusions:

• The DRUMBEAT algorithm provides a powerful tool for analyzing large MD datasets to understand protein dynamics.
• Identified specific residue communities crucial for D3R conformational transitions can inform the design of subtype-specific drugs.
• This approach holds potential for developing novel therapeutics for neuropsychiatric and substance use disorders.