INHERIT

INHERIT identifies bacteriophage genome sequences in metagenomic datasets to enable accurate phage detection and downstream analysis of phage composition and diversity.

Key Features:

• Integration of methods: Combines database-based (alignment-based) methods with alignment-free machine learning and deep learning approaches for sequence identification.
• Deep representation learning model: Employs a BERT-style deep representation learning framework for feature extraction and knowledge acquisition from sequence data.
• Pre-training strategy: Uses a pre-training strategy that separately processes two species to refine representation learning for non-alignment models.
• Performance and validation: Demonstrated superior identification performance versus four existing methods on a third-party benchmark, achieving an F1-score of 0.9932.

Scientific Applications:

• Metagenomic phage identification: Detects and classifies bacteriophage sequences within metagenomic datasets for surveys of viral diversity.
• Phage–microbe interaction studies: Supports investigation of bacteriophage roles in microbial ecology by providing accurate phage sequence calls.
• Phage therapy research: Provides sequence-level identification useful for applications exploring phage-based therapeutics.

Methodology:

Pre-training on existing sequence databases, separate pre-training for two species, and application of a BERT-style deep learning framework combined with database-based (alignment) and alignment-free machine learning approaches.