EC-Design: A Robust Framework for Enzyme Function Prediction Using Dimensionality-Reduced Sequence Features.

Area of Science:

• Bioinformatics
• Machine Learning
• Enzyme Catalysis

Background:

• Automated enzyme classification faces challenges due to high-dimensional data and imbalanced datasets.
• Existing methods often struggle with accuracy and interpretability in large-scale enzyme annotation.

Purpose of the Study:

• To develop an efficient and accurate machine learning framework, EC-Design, for automated enzyme classification.
• To evaluate the performance of instance-based learning against ensemble methods for enzyme annotation.

Main Methods:

• Utilized Principal Component Analysis (PCA) and Fisher Score for feature selection on 134,153 enzyme sequences.
• Implemented and compared k-nearest neighbors (k-NN) against six other machine learning algorithms.
• Analyzed model generalization, discriminative power (AUC), and class-specific performance metrics.

Main Results:

• k-nearest neighbors (k-NN) achieved a top accuracy of 74.59% and a macro-F1 score of 0.6859, outperforming ensemble methods.
• The model demonstrated robust generalization (74.37 ± 0.49%) and excellent discriminative power (mean AUC = 0.937).
• Identified dipeptide patterns containing asparagine and glycine as key discriminative features, offering biological interpretability.

Conclusions:

• EC-Design framework establishes instance-based learning as an effective approach for large-scale enzyme annotation.
• The study provides a transparent and accurate alternative to complex ensemble models for enzyme classification.
• Identified key sequence features that provide biologically interpretable insights into enzyme function.