Drug-Protein Interactions Prediction Models Using Feature Selection and Classification Techniques.

Area of Science:

• Computational chemistry and cheminformatics
• Bioinformatics and computational biology
• Machine learning in drug discovery

Background:

• Drug-Protein Interaction (DPI) identification is critical for drug discovery, but high-dimensional data presents challenges.
• Existing computational methods like docking-based and ligand-based approaches have limitations.
• Chemogenomics-based machine learning approaches offer a promising solution by integrating drug and protein features.

Purpose of the Study:

• To improve the accuracy and efficiency of drug-protein interaction (DPI) prediction.
• To address challenges posed by high dimensionality and data imbalance in DPI datasets.
• To identify optimal machine learning strategies for DPI prediction.

Main Methods:

• Utilized protein data from KEGG and drug data from DrugBank.
• Applied and evaluated various data balancing techniques: Random Over Sampling (ROS), SMOTE, and Adaptive SMOTE.
• Assessed feature selection methods including Correlation, Information Gain (IG), Chi-Square (CS), and Relief.
• Employed and compared classification algorithms: Support Vector Machines (SVM), Random Forest (RF), Adaboost, and Logistic Regression (LR).

Main Results:

• Evaluated the effectiveness of different balancing techniques to handle imbalanced Drug Protein Pairs (DPP).
• Compared multiple feature selection methods to identify the most informative drug and protein features.
• Determined the optimal combination of balancing, feature selection, and classification methods for accurate DPI prediction.

Conclusions:

• The study successfully identified the most effective combination of machine learning techniques for DPI prediction.
• This optimized approach enhances the reliability and efficiency of computational drug-protein interaction studies.
• The findings contribute to more accurate predictions in the field of drug discovery and development.