sketching-based

sketching-based applies sampling and sketching methods to rapidly classify long sequencing reads by identifying their source genomes for metagenomic read classification and contaminant detection.

Key Features:

• Sampling and Sketching Approaches: Generates a reduced representation or "screen" of potential source genomes to enable rapid read classification via similarity measures.
• Uniform Sampling: Uses uniform sampling across input sequences to select representative subsets for downstream sketching.
• MinHash and Variants: Employs MinHash, including weighted and order variants, to create compact sketches that capture essential genomic characteristics.
• Minimizer-Based Technique: Selects representative k-mers (minimizers) to construct minimal yet informative sketches.
• Clustering-Based Sketching Approach: Groups similar sequences to form more efficient representations for classification.
• Efficiency in Pre-processing: Requires storing and processing only subsets of input data, reducing time and space overheads relative to full alignment or indexing.
• High Accuracy: Maintains high read classification accuracy despite reduced representations by predicting read source based on similarity to sketch elements.

Scientific Applications:

• Metagenomic species identification: Identifies species origin of reads from complex microbial communities using sketch-based similarity.
• Contaminant detection: Distinguishes target reads from contaminant reads in sequencing experiments.
• Long-read classification: Applies to classification of long sequencing reads, including those with high error rates.

Methodology:

Methods explicitly include uniform sampling, MinHash (including weighted and order variants), a minimizer-based k-mer selection technique, and a clustering-based sketching approach.