Native or Non-Native Protein-Protein Docking Models? Molecular Dynamics to the Rescue

Area of Science:

• Computational Biology
• Structural Bioinformatics
• Biophysics

Background:

• Molecular docking generates numerous protein-protein complex models.
• Differentiating native-like models from decoys is a significant challenge in structural biology.

Purpose of the Study:

• To evaluate molecular dynamics (MD) simulations combined with machine learning for distinguishing native from non-native protein-protein complex models.
• To complement existing docking scoring functions with a simulation-based approach.

Main Methods:

• Generated initial models for 25 protein-protein complexes using HADDOCK.
• Employed MD simulations and machine learning (random forest classifier) to analyze model stability.
• Utilized metrics like root mean square deviation and fraction of native contacts for discrimination.

Main Results:

• Native models exhibited significantly higher stability across multiple measured properties compared to non-native models.
• The random forest classifier achieved an accuracy of 0.85 in distinguishing native from non-native complexes.
• Effective discrimination was achieved with relatively short simulation lengths (50-100 ns).

Conclusions:

• MD simulations coupled with machine learning provide a robust method for validating protein-protein docking models.
• This approach enhances the reliability of predicted protein-protein interactions.
• The protocol is practical for routine use in structural bioinformatics research.