HVSeeker: a deep-learning-based method for identification of host and viral DNA sequences.

Area of Science:

• Bioinformatics
• Microbiology
• Genomics

Background:

• Bacteriophages are abundant and vital to ecosystems, but identifying novel viral sequences in metagenomes is challenging due to rapid evolution.
• Traditional methods struggle with short or novel viral genomes, necessitating advanced computational approaches.

Purpose of the Study:

• To develop a deep learning-based method for accurate identification of viral sequences within mixed metagenomic samples.
• To improve the detection of novel and previously uncharacterized viral genomes.

Main Methods:

• Developed HVSeeker, a deep learning model with separate components for DNA and protein sequence analysis.
• Implemented three preprocessing techniques: padding, contigs assembly, and sliding window, to enhance sequence learning.
• Trained and tested HVSeeker on diverse datasets, including NCBI and IMGVR databases.

Main Results:

• HVSeeker demonstrated robust performance across various sequence lengths (200-1500 bp).
• Outperformed existing methods like Seeker, Rnn-VirSeeker, DeepVirFinder, and PPR-Meta in benchmark tests.
• Successfully identified unknown phage genomes, highlighting its effectiveness in detecting novel viral sequences.

Conclusions:

• HVSeeker's architecture, including preprocessing and model design, offers significant advancements in viral genome identification.
• The tool is crucial for prokaryotic and phage taxonomy and facilitates downstream analyses, including phage therapy development.
• Provides a critical first step in analyzing host-viral interactions within complex metagenomic samples.