EC-Bench: a benchmark for enzyme commission number prediction.

Area of Science:

• Biochemistry
• Bioinformatics
• Computational Biology

Background:

• Enzymes are crucial proteins catalyzing biochemical reactions.
• Enzyme Commission (EC) numbers classify enzymes based on their catalytic activity.
• Accurate EC number prediction is vital for understanding enzyme function.
• Existing methods for EC number prediction lack a unified evaluation framework.

Purpose of the Study:

• To introduce EC-Bench, a comprehensive benchmark for evaluating enzyme EC number prediction methods.
• To provide a standardized framework for objective comparison of diverse prediction approaches.
• To facilitate the introduction and evaluation of novel enzyme annotation methods.

Main Methods:

• EC-Bench includes a representative set of existing prediction methods (homology-based, deep learning, contrastive learning, language models).
• It incorporates established and novel performance metrics for accuracy and efficiency.
• The benchmark utilizes selected datasets for comprehensive comparative studies.

Main Results:

• Extensive experiments were conducted to compare existing EC number prediction methods.
• Performance variations were observed across different methods and prediction tasks (exact prediction, completion, recommendation).
• Subtle yet potentially useful performance differences were identified for various methods across EC hierarchy levels.

Conclusions:

• EC-Bench offers an open-source, unified framework for evaluating enzyme EC number prediction methods.
• The benchmark enables objective comparison of methods under uniform conditions.
• EC-Bench facilitates the identification of the most effective enzyme annotation strategies and aids in understanding method-specific strengths and weaknesses.