IterCluster

IterCluster performs barcode clustering to group long fragment read (LFR) barcodes for improved haplotyping and de novo genome assembly using data from single tube long fragment reads (stLFR), 10X Genomics Chromium reads, and TruSeq synthetic long-reads.

Key Features:

• Alignment-free clustering: Uses an alignment-free clustering approach to group barcodes without read mapping.
• -mer frequency features: Represents barcodes using -mer frequency-based features computed from barcode-associated reads.
• Markov Cluster (MCL) algorithm: Applies the Markov Cluster (MCL) algorithm to cluster barcodes based on feature similarity.
• Divide-and-conquer strategy: Employs a divide-and-conquer strategy to generate clusters that increase sequencing depth per target region.
• Sparse coverage handling: Addresses sparse read coverage per barcode and multiple genomic-region assignments per barcode to improve cluster quality.
• Assembly metric improvement: Reported to improve scaffold and contig N50 in human genome datasets (from 13.2 kbp/7.1 kbp to 17.1 kbp/11.9 kbp in the cited example).
• Compatible LFR technologies: Applicable to BGI stLFR, 10X Genomics Chromium reads, and TruSeq synthetic long-reads.
• Improved clustering performance: Demonstrated superior clustering precision and recall on BGI stLFR data compared to 10X Genomics Chromium datasets in the reported evaluation.

Scientific Applications:

• Haplotyping: Groups LFR barcodes to support phasing and haplotype reconstruction using long-range barcode information.
• De novo genome assembly: Enhances de novo assembly by increasing effective coverage within barcode clusters to improve contig and scaffold contiguity.
• LFR barcode enrichment assessment: Identifies barcode sets enriched for specific genomic target regions to support downstream assembly or analysis.
• Cross-technology LFR analysis: Enables comparative and integrative analyses across stLFR, 10X Genomics Chromium, and TruSeq synthetic long-read datasets.

Methodology:

Computes -mer frequency-based features from barcode-associated reads, applies an alignment-free clustering approach using the Markov Cluster (MCL) algorithm, and refines clusters with a divide-and-conquer strategy.