EC2Vec: A Machine Learning Method to Embed Enzyme Commission (EC) Numbers into Vector Representations.

Area of Science:

• Bioinformatics
• Enzymology
• Machine Learning

Background:

• Enzyme Commission (EC) numbers are crucial for enzyme classification and function understanding.
• Existing EC number encoding methods for machine learning face challenges like false numerical ordering and high data sparsity.
• Effective encoding is vital for advancing enzyme-related computational research.

Purpose of the Study:

• To develop an advanced encoding method for Enzyme Commission (EC) numbers that overcomes limitations of current approaches.
• To create informative and meaningful enzyme representations by preserving categorical nature and hierarchical structure.
• To enhance the performance of machine learning models in enzyme-related research.

Main Methods:

• Developed EC2Vec, a multimodal autoencoder for encoding EC numbers.
• Treated each digit of the EC number as a categorical token.
• Utilized a 1D convolutional layer to process embeddings and capture relationships.
• Benchmarked EC2Vec against existing simple encoding methods using a large EC number dataset.

Main Results:

• EC2Vec effectively preserves the categorical nature and hierarchical structure of EC numbers.
• t-SNE visualization showed distinct clusters for different enzyme classes, confirming hierarchical capture.
• EC2Vec embeddings demonstrated superior performance over other methods in reaction-EC pair classification.
• Benchmarking confirmed EC2Vec's advantage over simple encoding techniques.

Conclusions:

• EC2Vec provides robust and informative representations of EC numbers, leveraging their inherent hierarchy.
• The method significantly improves downstream machine learning tasks, such as enzyme classification.
• EC2Vec offers a valuable tool for enzyme research and bioinformatics applications.