Empathi: embedding-based phage protein annotation tool by hierarchical assignment

Area of Science:

• Microbiology
• Bioinformatics
• Virology

Background:

• Bacteriophages are abundant viruses crucial to microbial ecosystems.
• Rapid evolution of bacteriophages results in many proteins with unknown functions.
• Existing methods for phage protein function identification lack sensitivity or specificity.

Purpose of the Study:

• To introduce Empathi, a novel protein-embedding-based classifier for hierarchical phage protein function assignment.
• To develop new, hierarchically organized functional categories for phage proteins suitable for machine learning.

Main Methods:

• Empathi utilizes protein embeddings for function classification.
• Hierarchical functional categories were specifically designed for phage proteins.
• Performance was evaluated on cultured phage genomes and the EnVhogDB database.

Main Results:

• Empathi significantly outperforms homology-based methods in annotating phage protein functions.
• It tripled the number of annotated homologous groups on cultured phage genomes.
• Empathi doubled the annotated fraction of protein families in the EnVhogDB database (from 16% to 33%).

Conclusions:

• Empathi provides a more sensitive and specific method for identifying phage protein functions.
• Improved functional annotation enhances understanding of phage biology and their interactions with bacteria.
• This tool aids in deciphering the functional repertoire of phages in various ecosystems.