CONSENT

CONSENT performs scalable long-read self-correction and assembly polishing by combining multiple sequence alignment and local de Bruijn graph polishing to reduce ~10% error rates in Oxford Nanopore Technologies and other third-generation sequencing long reads.

Key Features:

• Segmentation-based overlap computation: Computes overlaps between long reads using a segmentation strategy to form alignment piles for each read.
• Windowed multiple sequence alignment: Divides alignment piles into smaller windows and derives consensus sequences by performing multiple sequence alignment within each window.
• Local de Bruijn graph polishing: Applies local de Bruijn graphs to polish consensus sequences and reduce residual errors.
• Assembly polishing: Polishes raw assemblies directly to improve assembly accuracy.
• Ultra-long read scalability: Scales to ultra-long reads (reported up to 1.5 Mbp) and has been applied to large human datasets (processing example: 1.5 Mbp reads in ~10 days).
• Comparative performance: Reported to be 2–38 times faster than other polishing tools on real Oxford Nanopore data while delivering comparable or superior results and improving assemblies produced by Flye.
• Resource-efficient workflow: Demonstrates that assembling raw reads and then polishing the assembly can be less resource-intensive and yield better results than correcting reads prior to assembly.

Scientific Applications:

• Genome assembly improvement: Improves the accuracy of long-read assemblies for downstream genomic analyses.
• Structural variant detection: Increases read and assembly accuracy to support more reliable structural variant identification.
• Complex region characterization: Enables analysis of complex genomic regions by improving continuity and correctness of ultra-long reads and assemblies.
• Long-read sequencing data refinement: Reduces error rates in third-generation sequencing datasets, facilitating more accurate downstream interpretations.

Methodology:

Compute overlaps between long reads using a segmentation strategy to form alignment piles, split piles into windows, derive consensus sequences by multiple sequence alignment within windows, and polish these consensuses using local de Bruijn graphs.