Predicting Conserved Water Molecules in Binding Sites of Proteins Using Machine Learning Methods and Combining...

Area of Science:

• Computational Biology
• Structural Bioinformatics
• Machine Learning in Biochemistry

Background:

• Water molecules are crucial for protein structure, folding, and binding affinity.
• Distinguishing conserved water molecules (CWMs) from free water molecules (FWMs) is difficult due to their embedded nature and strong hydrogen bonds.
• Environmental factors complicate direct identification of CWMs.

Purpose of the Study:

• To develop a machine learning framework for identifying CWMs in protein binding sites.
• To leverage spatial structure and physicochemical properties of water molecules for CWM identification.
• To compare the performance of various machine learning models for this task.

Main Methods:

• Extracted six key features: atom density, hydrophilicity, hydrophobicity, solvent-accessible surface area, temperature B-factors, and mobility.
• Analyzed and combined features to optimize CWM identification rates.
• Evaluated seven machine learning models (SVM, KNN, DT, LR, DA, NB, EL) using the optimal feature combination.

Main Results:

• An optimal feature combination was determined for enhanced CWM identification.
• Ensemble learning (EL) model demonstrated superior performance compared to other models.
• The EL model achieved satisfactory prediction accuracy in identifying CWMs and FWMs.

Conclusions:

• The developed machine learning framework effectively identifies conserved water molecules in protein binding sites.
• The ensemble learning model, combined with optimal feature selection, provides a robust method for CWM prediction.
• Validated through case studies and comparison with existing tools like Dowser++.