winnowmap

winnowmap maps Oxford Nanopore Technologies (ONT) and Pacific Biosciences (PacBio) long reads to reference genomes with improved accuracy in repetitive genomic regions using weighted-minimizer sampling.

Key Features:

• Minimap2 codebase: Built on the Minimap2 codebase to leverage its long-read alignment framework.
• Weighted-minimizer sampling algorithm: Assigns weights to k-mers to down-weight frequent k-mers while maintaining matching minimizers for shared substrings to improve mapping precision in repeats.
• Improved accuracy in repetitive regions: Enhances alignment in segmental duplications and tandem repeat arrays for ONT and PacBio long reads.
• Confident subalignment strategy (Winnowmap2): Uses minimal confidently alignable substrings to compute mappings via collections of confident subalignments, increasing tolerance to structural variants and sensitivity to paralog-specific variants and reducing allelic bias.
• Reduction in mapping error rates: Empirical evaluations report a decrease in mapping error rate on the human X chromosome from 0.14% to 0.06% and on the X centromere from 3.6% to approximately 0%.
• Index compression and runtimes: Sparser sampling enables improved index compression and contributes to competitive runtimes for large-scale analyses.

Scientific Applications:

• Variant and structural variation analysis: Facilitates more reliable variant calling and structural variation detection in repetitive genomic regions.
• Paralog-specific and allelic bias studies: Reduces allelic bias and increases sensitivity to paralog-specific variants, aiding analyses of paralogous genes and highly similar repeat copies.
• Long-read sequencing projects: Applicable to analyses of ONT and PacBio long-read datasets that require accurate mapping through repeats.

Methodology:

Implements weighted-minimizer sampling with k-mer weighting and, in Winnowmap2, minimal confidently alignable substrings for confident subalignments, built on the Minimap2 codebase.