Unveiling Switching Function of Amino Acids in Proteins Using a Machine Learning Approach

Area of Science:

• Biophysics
• Computational Biology
• Protein Science

Background:

• Protein structure and function are dictated by amino acid dynamics.
• Current techniques limit understanding of residue-level protein structural features.
• Investigating amino acid dynamics is crucial for comprehending protein behavior.

Purpose of the Study:

• To develop a novel machine-learning (ML) framework for analyzing individual amino acid dynamics.
• To identify major conformational states and classify amino acids based on their switching modes.
• To evaluate the dynamic stability of amino acid residues within protein structures.

Main Methods:

• Utilized Molecular Dynamics (MD) trajectories as input for the ML framework.
• Employed a Random Forest model for classifying amino acid switching behaviors.
• Developed criteria to distinguish between stable switch (SS) and unstable switch (US) residues.

Main Results:

• Achieved 96.94% classification accuracy in identifying switch residues using the Random Forest model.
• Successfully differentiated between stable switch (SS) and unstable switch (US) amino acid residues.
• Demonstrated a correlation between the dynamics of SS residues and global protein properties.

Conclusions:

• The novel ML framework effectively analyzes amino acid dynamics at the residue level.
• The classification of SS and US residues provides new insights into protein conformational flexibility.
• Understanding residue-level dynamics is key to predicting and understanding protein global properties.