DeePhage

DeePhage classifies metavirome DNA sequences into temperate phage-derived and virulent phage-derived categories to distinguish phage lifestyles in metagenomic studies.

Key Features:

• Phage lifestyle classification: Classifies metavirome sequences into temperate phage-derived and virulent phage-derived categories.
• Deep learning (convolutional neural network): Applies a convolutional neural network to DNA sequences to extract local sequence features for classification.
• One-hot encoding representation: Represents DNA sequences using one-hot encoding as nucleotide-level input to the neural network.
• Performance metrics: Achieves up to 88% accuracy in five-fold validation, approximately 30% higher than PHACTS, and annotates 54.4% of reliable contigs versus 44.5% by PHACTS.
• Computational efficiency: Reduces processing time by 810-fold compared with PHACTS on the same machine.
• Direct detection of temperate viral fragments: Identifies temperate viral fragments from metagenome and metavirome data to investigate phage transformations.

Scientific Applications:

• Microbial ecology and phage-host interactions: Enables analysis of phage roles and dynamics within microbial communities by distinguishing temperate and virulent phages.
• Phage transformation studies: Facilitates investigation of temperate-to-virulent transformations by detecting temperate viral fragments in metagenomic datasets.
• Therapeutic implications: Provides insights that may inform exploration of potential therapeutic applications related to phage transformations in human disease contexts.
• Metavirome sequence annotation: Increases the proportion of annotated reliable contigs in metavirome datasets compared with PHACTS.

Methodology:

One-hot encoded DNA sequences are input to a convolutional neural network that extracts local sequence features; model performance was evaluated by five-fold cross-validation.