Recognizing Ion Ligand-Binding Residues by Random Forest Algorithm Based on Optimized Dihedral Angle.

Area of Science:

• Computational Biology
• Biochemistry
• Bioinformatics

Background:

• Predicting ion ligand-binding residues is crucial for understanding protein function in biological processes.
• Identifying these residues aids in deciphering protein-ligand interactions and their roles.

Purpose of the Study:

• To develop and validate a robust prediction model for identifying ion ligand-binding residues in protein sequences.
• To improve upon existing methods for predicting binding sites of metal and acid radical ions.

Main Methods:

• Utilized amino acid sequence composition, conservation, predicted structural information, and physicochemical properties as features.
• Developed novel feature extraction techniques using information entropy for polarization and hydrophobicity, and position weight matrices for conservation.
• Employed the random forest algorithm for the prediction model, incorporating classification and combination of phi and psi dihedral angles.

Main Results:

• The random forest model demonstrated superior prediction performance compared to previous studies.
• Achieved Matthew's correlation coefficient > 0.07 and accuracy > 52% for 10 metal ion ligands.
• Achieved Matthew's correlation coefficient > 0.15 and accuracy > 86% for four acid radical ion ligands.

Conclusions:

• The proposed methods significantly enhance the accuracy of predicting ion ligand-binding residues.
• The optimized model provides a valuable tool for studying protein-ligand interactions and protein function.
• Further refinement by classifying and combining dihedral angles improved model performance for specific ion ligands.