FALCON2: compression-based metagenomic classification of ancient viruses

Area of Science:

• Paleogenomics
• Bioinformatics
• Computational Biology

Background:

• Ancient DNA (aDNA) presents significant challenges for taxonomic classification due to fragmentation, deamination, and contamination.
• Conventional metagenomic classifiers struggle with short, damaged aDNA reads, limiting paleogenomic analyses.

Purpose of the Study:

• To develop and present FALCON2, a compression-based metagenomic classifier designed for high accuracy on degraded aDNA.
• To improve upon existing methods for taxonomic classification of ancient viral sequences.

Main Methods:

• Leverages position-aware finite-context models for classification.
• Consolidates FALCON-meta capabilities into a unified executable with enhanced features.
• Includes model persistence, direct compressed input processing, and optional pre-filtering.

Main Results:

• FALCON2 achieved superior performance on simulated viral datasets, with an AUC-ROC of 0.999, AUPRC of 0.968, and F1-score of 0.918.
• Outperformed Centrifuge, Kraken2, and CLARK-S, particularly on ultra-short reads (20-40 bp).
• Pre-filtering improved precision by 10 percentage points with minimal recall loss.

Conclusions:

• FALCON2 offers a robust solution for taxonomic classification of challenging ancient DNA samples.
• The tool enhances paleogenomic research by enabling accurate analysis of degraded and contaminated sequences.
• FALCON2 is efficient, requiring 4-8 GB RAM for typical analyses and is freely available.