DeePVP: Identification and classification of phage virion proteins using deep learning.

Area of Science:

• Bioinformatics
• Virology
• Genomics

Background:

• Phage virion proteins (PVPs) determine crucial biological properties.
• Poor annotation of PVPs hinders viral research areas like phylogenetic analysis and host identification.
• High sequence diversity makes phage genome PVP annotation a challenging bioinformatic task.

Purpose of the Study:

• To develop a deep learning-based tool, DeePVP, for accurate phage virion protein (PVP) annotation.
• To discriminate PVPs from non-PVPs and classify identified PVPs into major classes.
• To provide a reliable tool for analyzing phage genomic structures and identifying novel PVP-enriched regions.

Main Methods:

• Developed DeePVP, a deep learning model with a main module for PVP identification and an extended module for PVP classification.
• Evaluated DeePVP's performance against state-of-the-art tools for PVP identification and classification tasks.
• Utilized virtual machine with GUI and Docker for optimized program execution, ensuring accessibility for non-computer professionals.

Main Results:

• DeePVP's main module achieved a 9.05% higher F1-score in PVP identification compared to existing tools.
• DeePVP's extended module demonstrated approximately 3.72% higher accuracy in PVP classification than PhANNs.
• Application cases confirmed DeePVP's reliability in revealing compact PVP-enriched regions, including a novel discovery in Escherichia phage phiEC1.

Conclusions:

• DeePVP outperforms current state-of-the-art tools in PVP annotation.
• The tool is optimized for ease of use via GUI and Docker, making it accessible to a wider range of researchers.
• DeePVP is a valuable addition to phage genomic analysis, facilitating discoveries in viral structures and functions.