Prediction of enzymatic function with high efficiency and a reduced number of features using genetic algorithm

Area of Science:

• Bioinformatics
• Computational Biology
• Machine Learning

Background:

• The post-genomic era necessitates efficient methods for protein function identification.
• Machine learning approaches, particularly feature-based methods, are crucial in bioinformatics.
• Understanding protein structures (primary, secondary, tertiary, quaternary) is key to function prediction.

Purpose of the Study:

• To investigate protein characteristics that enhance machine learning model quality for enzyme class prediction.
• To apply dimensionality reduction techniques and Support Vector Machine (SVM) classification.
• To develop and evaluate feature selection methods, including a novel genetic algorithm approach.

Main Methods:

• Utilized Factor Analysis for feature extraction/transformation.
• Developed a multi-objective genetic algorithm for feature selection.
• Compared the proposed genetic algorithm with other feature selection methods.
• Employed Support Vector Machine (SVM) for enzyme classification.

Main Results:

• A feature subset generated by the multi-objective genetic algorithm reduced the dataset by approximately 87%.
• This subset achieved an F-measure of 85.78%, significantly improving classification model quality.
• A reduced set of 28 features achieved over 80% F-measure for four out of six enzyme classes.

Conclusions:

• Satisfactory enzyme classification performance can be achieved using a significantly reduced set of protein characteristics.
• The developed multi-objective genetic algorithm is effective for selecting relevant features in bioinformatics.
• Openly available datasets and implementations facilitate further research in protein function prediction.