DeepHost

DeepHost predicts phage host taxonomy from next-generation sequencing (NGS) phage genomes using machine learning to identify host organisms.

Key Features:

• Convolutional Neural Network (CNN) Architecture: DeepHost uses a convolutional neural network (CNN) to infer host taxonomies from phage genome sequences.
• Genome Encoding Methodology: Genomes are encoded as matrices using spaced k-mer pairs to represent sequences, enabling robustness to insertions, deletions, and mutations.
• High Prediction Accuracy: Achieves genus-level accuracy of 96.05% across 72 taxonomies and species-level accuracy of 90.78% across 118 taxonomies, outperforming existing tools by 10.16%–30.48% and achieving performance comparable to BLAST.
• Alignment-Free and Efficient: Operates in an alignment-free manner to enable faster processing of large datasets.
• Applicability to Diverse Genomic Data: Provides predictions for sequences without BLAST hits, with 38.00% genus-level accuracy and 26.47% species-level accuracy.

Scientific Applications:

• Phage-host interaction inference: Inferring phage-host relationships from NGS-derived phage genomes.
• Microbial ecology: Characterizing viral components of microbial communities to study ecology and community dynamics.
• Virology and viral diversity: Assigning hosts to uncharacterized phages to support studies of viral diversity and evolution.

Methodology:

Convolutional neural network trained on genome encodings represented as matrices of spaced k-mer pairs; alignment-free prediction; encoding designed to handle insertions, deletions, and mutations.