evERdock BAI: Machine-learning-guided selection of protein-protein complex structure

Area of Science:

• Computational biology
• Structural bioinformatics
• Biophysics

Background:

• Protein-protein interactions are crucial in biological systems.
• Accurate prediction of protein-protein complex structures aids in understanding these interactions.
• Existing methods like evERdock, while effective, are computationally expensive.

Purpose of the Study:

• To develop a more computationally efficient method for selecting accurate protein-protein complex structures.
• To reduce the computational cost associated with high-accuracy decoy selection.

Main Methods:

• Integration of the evERdock method with the best arm identification (BAI) framework from reinforcement learning.
• BAI framework optimizes calculations by prioritizing promising decoys and reducing computations for non-promising ones.
• Evaluation on three protein-protein complex systems.

Main Results:

• The proposed method successfully reduced computational costs compared to standard approaches.
• Computational cost reduction achieved up to a factor of 4.05 in the best case.
• Prediction accuracy was maintained without sacrifice.

Conclusions:

• The combination of evERdock and BAI offers a significant improvement in computational efficiency for protein-protein complex structure prediction.
• This approach provides a viable solution for reducing the computational burden in structural bioinformatics.
• The method demonstrates the potential of reinforcement learning in optimizing complex biological simulations.