Neural relational inference to learn long-range allosteric interactions in proteins from molecular dynamics...

Area of Science:

• Biophysics
• Computational Biology
• Structural Biology

Background:

• Protein allostery involves long-range communication within proteins, crucial for biological regulation.
• Molecular dynamics (MD) simulations are valuable for studying allostery but are limited by timescale.
• Deep learning offers new capabilities for analyzing complex protein dynamics.

Purpose of the Study:

• To develop and apply a deep learning model for inferring allosteric communication pathways.
• To investigate protein allostery using a novel graph neural network-based approach.
• To enhance the analysis of dynamic residue interactions in allosteric processes.

Main Methods:

• Applied a neural relational inference model with an encoder-decoder architecture.
• Utilized graph neural networks to model proteins as dynamic networks of interacting residues.
• Analyzed MD simulation trajectories of Pin1, SOD1, and MEK1 systems.

Main Results:

• Successfully learned long-range interactions mediating allosteric communication.
• Identified communication pathways between distant sites in studied protein systems.
• Discovered allostery-related interactions earlier in simulations and improved mutation effect predictions.

Conclusions:

• Deep learning models can effectively probe protein allostery and uncover communication mechanisms.
• The developed model advances the computational analysis of dynamic protein networks.
• This method offers a more accurate and efficient way to study allosteric effects and predict mutation impacts.