CHEER

CHEER performs hierarchical, read-level taxonomic classification of RNA virus sequencing reads from order to genus using k-mer embedding and hierarchically organized convolutional neural networks.

Key Features:

• Hierarchical classification: Performs read-level taxonomic assignment for RNA viruses from order down to genus, including reads from previously uncharacterized species.
• K-mer embedding encoding: Encodes input reads using k-mer embedding-based representations.
• Hierarchically organized CNNs: Uses hierarchically organized convolutional neural networks to model taxonomic relationships at multiple taxonomic levels.
• Rejection layer: Incorporates a trained rejection layer to filter out non-target or ambiguous reads.
• Deep learning of species-specific features: Applies deep learning to learn species-specific features that enable detection of novel taxa beyond alignment-based methods.
• Comparison to alignment methods: Demonstrates higher taxonomic assignment accuracy than traditional alignment-based and alignment-free methods on benchmark datasets.
• Validation datasets: Validated on both simulated and real sequencing datasets, including short-read viral metagenomic data.

Scientific Applications:

• Viral metagenomics classification: Taxonomic classification of RNA viruses in viral metagenomic datasets.
• Novel virus detection: Detection and taxonomic assignment of previously uncharacterized or novel RNA viral species.
• Read-level taxonomic analysis: Assigning taxonomic labels to individual sequencing reads in short-read metagenomic datasets.
• Comparative benchmarking: Comparative performance assessment against alignment-based and alignment-free taxonomic classification methods.

Methodology:

CHEER encodes reads using k-mer embedding-based representations, applies hierarchically organized convolutional neural networks with a trained rejection layer, and uses deep learning to learn species-specific features for read-level classification from order to genus.