vHULK

vHULK predicts bacteriophage hosts at genus and species levels from phage genomic features using deep neural networks to support phage research and phage therapy.

Key Features:

• Genomic Feature Utilization: Leverages 9,504 annotated genomic features derived from phage genomes based on alignment significance scores between predicted protein sequences and a curated database of viral protein families.
• Deep Neural Network Integration: Employs four deep neural network models trained to predict hosts at both genus and species levels targeting 61 host genera and 52 host species.
• High Accuracy Performance: Achieves 99% genus-level and 98% species-level accuracy on controlled test sets and mean accuracies of 82% (genus) and 52% (species) on a 2,178-phage-genome validation dataset.
• Superior Performance: Outperforms other existing phage host prediction programs on the same validation dataset.

Scientific Applications:

• Phage Therapy: Predicts bacteriophage hosts to identify candidate phages for therapeutic applications against bacterial pathogens.
• Microbial Ecology and Evolution: Supports studies of the ecological roles of bacteriophages and their evolutionary relationships with bacterial hosts.

Methodology:

Predicted protein sequences from phage genomes are aligned to a curated viral protein family database to produce alignment significance scores that form 9,504 annotated genomic features, which are input to four deep neural network models trained and evaluated (including a 2,178-genome validation set) for genus- and species-level host prediction.