Predicting network of drug-enzyme interaction based on machine learning method.

Area of Science:

• Computational biology
• Bioinformatics
• Drug discovery

Background:

• Accurate mapping of drug-enzyme interactions is vital for drug research.
• Existing methods may lack efficiency or comprehensive feature representation.

Purpose of the Study:

• To develop a novel computational approach for predicting drug-enzyme interactions.
• To identify key molecular features influencing these interactions.

Main Methods:

• Encoding drug molecules using physicochemical descriptors.
• Encoding enzyme molecules using pseudo amino acid composition.
• Utilizing Random Forest for network construction and prediction.
• Feature selection identifying 129 optimal features across nine categories.

Main Results:

• Geometry features identified as most significant predictors.
• Achieved high performance metrics: MCC of 0.915 (10-fold CV) and 0.895 (independent set).
• Demonstrated high sensitivity (87.9% CV, 95.7% independent) at high specificity levels (99.8% CV, 95.4% independent).

Conclusions:

• The proposed method effectively predicts drug-enzyme interactions.
• The feature set provides insights into the determinants of drug-enzyme binding.
• This approach has significant implications for computational proteomics and systems biology.