GPCR molecular dynamics forecasting using recurrent neural networks.

Area of Science:

• Biochemistry and structural biology
• Computational biophysics
• Pharmacology

Background:

• G protein-coupled receptors (GPCRs) are crucial cell membrane proteins mediating extracellular signals.
• Signal transduction involves conformational changes in the GPCR transmembrane region, necessitating dynamic studies.
• Molecular dynamics (MD) simulations offer detailed insights into biomolecular structure and function.

Purpose of the Study:

• To predict the dynamics of G protein-coupled receptors (GPCRs) in active and inactive states using machine learning.
• To analyze GPCR activation pathways and specific receptor regions under different ligand conditions.
• To evaluate and compare the performance of various neural network architectures for predicting protein dynamics.

Main Methods:

• Utilized Long Short-Term Memory (LSTM), a type of Recurrent Neural Network (RNN), for dynamic predictions.
• Performed MD simulations for two GPCR states across six scenarios, including APO and agonist/inverse agonist treatments.
• Evaluated four machine learning models with varying neural network complexity.

Main Results:

• The best performing model achieved a root-mean-square deviation (RMSD) below 0.139 Å.
• Transmembrane helices exhibited the lowest prediction errors and minimal relative movements.
• LSTM models demonstrated significant accuracy in predicting GPCR dynamics.

Conclusions:

• Machine learning, specifically LSTM networks, can effectively predict GPCR dynamics.
• The transmembrane regions are key areas for understanding GPCR conformational changes.
• This approach provides a powerful tool for investigating GPCR function and drug interactions.