Perseus: Lineage-Aware Refinement of Kraken2 Taxonomic Classification for Long Read Metagenomes.

Perseus reduces false taxonomic assignments by up to 50% in long-read metagenomic data through lineage-aware confidence modeling.

• Why it matters: Accurate taxonomic classification of long reads is crucial for understanding complex microbial communities, but current classifiers like Kraken2 often over-assign labels, especially with novel taxa, leading to high false positive rates.
• What they did: The authors developed Perseus, a framework that uses a multi-headed convolutional neural network to estimate confidence scores for taxonomic correctness at each rank, refining Kraken2 outputs by modeling spatial and hierarchical evidence.
• The result: Perseus consistently improves precision and lineage consistency, particularly for long reads and contigs, enabling more reliable microbial community analysis and reducing erroneous classifications.