An enhanced methodology for predicting protein-protein interactions between human and hepatitis C virus via ensemble...

Area of Science:

• Computational biology
• Virology
• Bioinformatics

Background:

• Hepatitis C virus (HCV) causes severe liver diseases like cirrhosis and hepatocellular carcinoma (HCC).
• Understanding human-HCV protein-protein interactions is crucial for developing effective antiviral therapies.
• Computational approaches can accelerate the discovery of new treatments for HCV infections.

Purpose of the Study:

• To construct a robust prediction model for human-HCV protein-protein interactions.
• To identify optimal features and machine learning algorithms for accurate prediction.
• To provide a valuable tool for future experimental research in HCV drug development.

Main Methods:

• Utilized pseudo amino acid compositions for feature generation at category and feature levels.
• Employed Extra-Tree for feature selection and Support Vector Machine (SVM) for initial classification model building.
• Compared and selected the best models using ensemble learning algorithms: Random Forest, Adaboost, and Xgboost.

Main Results:

• Profile-based features demonstrated superior performance in predictive model construction.
• The Xgboost model achieved an AUC of 92.66% on an independent dataset.
• Profile-based Physicochemical Distance Transformation, used with Adaboost, yielded the highest AUC of 93.74%.

Conclusions:

• A novel computational model with enhanced prediction accuracy for human-HCV protein-protein interactions was developed.
• The findings offer a practical reference for experimental investigations into HCV-related diseases.
• This study contributes to advancing HCV therapy and optimizing treatment strategies.