MOSAIK

MOSAIK maps second- and third-generation sequencing reads to reference genomes to produce accurate alignments for variant discovery and structural variant analysis.

Key Features:

• Multi-platform read support: Aligns reads from Illumina, Applied Biosystems SOLiD, Roche 454, Ion Torrent, and Pacific BioSciences SMRT.
• Hash clustering and Smith-Waterman: Employs a hash clustering strategy combined with the Smith-Waterman algorithm to capture mismatches and short insertions and deletions.
• Structural variant handling: Explicitly supports known-sequence structural variants such as mobile element insertions (MEIs) and generates outputs tailored for SV discovery.
• Neural-network mapping quality: Uses a neural-network-based training scheme to provide calibrated mapping quality scores with a reported correlation exceeding 0.98 between assigned and actual qualities.
• Genome-specific training pipeline: Includes a training pipeline to adapt mapping quality calibration for specific genomes.
• Multi-threaded execution: Implements multi-threading to improve computational efficiency.
• Benchmarking and validation: Validated across real and simulated datasets including the 1000 Genomes Project and supports benchmarking using the CuReSim read simulator.
• Workflow integration: Integrates into command and pipeline launcher systems such as GKNO.

Scientific Applications:

• Cross-platform read alignment: Mapping second- and third-generation sequencing reads to reference genomes from diverse sequencing technologies.
• Variant discovery: Detection and accurate alignment of mismatches and short indels to support SNP and small indel calling.
• Structural variant analysis: Discovery and characterization of structural variants, including mobile element insertions (MEIs).
• Mapper benchmarking: Evaluation and benchmarking of mapping algorithms using real and simulated high-throughput sequencing datasets.
• Mapping quality calibration: Generation of well-calibrated mapping quality scores for downstream genomic analyses.

Methodology:

Combines a hash clustering strategy with Smith-Waterman alignment, applies a neural-network-based training scheme for mapping quality calibration, supports multi-threaded execution, and uses benchmarking with real and simulated datasets including CuReSim and a genome-specific training pipeline.