Rectifying AI-generated protein structure ensembles for equilibrium using physics-based computations

Area of Science:

• Computational Biology
• Structural Biology
• Artificial Intelligence in Biochemistry

Background:

• Recent advancements in artificial intelligence (AI) have led to the development of tools capable of generating protein structure ensembles.
• These AI-generated ensembles often exhibit significant structural diversity and inconsistencies.

Purpose of the Study:

• To develop and validate a computational method for harmonizing diverse AI-generated protein structure ensembles.
• To achieve a consistent, atomically-detailed equilibrium ensemble of protein conformations.

Main Methods:

• Utilized a two-stage physics-based approach combining weighted ensemble (WE) simulations and the RiteWeight (RW) algorithm.
• AI-generated ensembles were used to seed WE simulations for relaxation toward a steady state.
• Trajectory segments from WE simulations were reweighted to the steady state using the RW algorithm.

Main Results:

• Successfully harmonized distinct AI-generated ensembles into a consistent equilibrium description.
• Demonstrated the effectiveness of the WE-RW approach in reconciling structural variations from different AI tools.
• Generated an atomically-detailed equilibrium ensemble for unliganded adenylate kinase.

Conclusions:

• The WE-RW computational path effectively unifies diverse AI protein structure predictions.
• This method provides a reliable approach to obtain consistent equilibrium ensembles, irrespective of the initial AI tool.
• Enables a more accurate and unified understanding of protein conformational dynamics.