Source-linked AI summary
SAMBLASTER: fast duplicate marking and structural variant read extraction
Gregory G. Faust, Ira M. Hall
TL;DR
Growing sequencing data make repeated BAM processing and structural-variant read extraction costly pipeline steps. SAMBLASTER performs duplicate marking and extracts relevant reads during a piped alignment post-pass, reducing intermediate processing while supporting downstream analyses.
Problem
Growing sequencing datasets require duplicate removal, while repeated BAM processing and extracting informative structural-variant reads add costly pipeline steps.
Method
SAMBLASTER marks duplicates in one pass over read-id-sorted SAM and simultaneously extracts discordant, split-read, unmapped, or clipped mappings.
Results
The piped post-pass avoids intermediate-file processing cycles and produces substantially smaller structural-variant input files from aligner output.
Takeaways & Limitations
SAMBLASTER integrates duplicate marking and structural-variant read extraction into existing alignment pipelines to improve processing efficiency.
Abstract
from arXiv · showhide
Motivation: Illumina DNA sequencing is now the predominant source of raw genomic data, and data volumes are growing rapidly. Bioinformatic analysis pipelines are having trouble keeping pace. A common bottleneck in such pipelines is the requirement to read, write, sort and compress large BAM files multiple times. Results: We present SAMBLASTER, a tool that reduces the number of times such costly operations are performed. SAMBLASTER is designed to mark duplicates in read-sorted SAM files as a piped post-pass on DNA aligner output before it is compressed to BAM. In addition, it can simultaneously output into separate files the discordant read-pairs and/or split-read mappings used for structural variant calling. As an alignment post-pass, its own runtime overhead is negligible, while dramatically reducing overall pipeline complexity and runtime. As a stand-alone duplicate marking tool, it performs significantly better than PICARD or SAMBAMBA in terms of both speed and memory usage, while achieving nearly identical results. Availability: SAMBLASTER is open source C++ code and freely available from https://github.com/GregoryFaust/samblaster
1 FEATURES AND METHODS
SAMBLASTER is a single-pass, read-id-sorted SAM post-pass that marks duplicates while supporting efficient pipeline streaming and structural-variant read extraction. Its methods include secondary-alignment propagation and a compact signature-based data structure designed to reduce memory use.
- Duplicate marking: SAMBLASTER marks duplicates in a single pass over read-id-sorted SAM, enabling aligner output to be piped directly into downstream SAM/BAM tools.This avoids storing the input file for a second pass.
- Duplicate marking: SAMBLASTER marks as duplicate any secondary alignment associated with a duplicate primary alignment, a capability absent from PICARD and SAMBAMBA at writing.The feature is particularly useful with BWA-MEM output.
- Structural-variant extraction: SAMBLASTER extracts discordant read-pairs and split-read mappings directly from aligner SAM output during an existing full-dataset read.These mappings are typically more than 100-fold smaller than the full dataset, improving pipeline efficiency.
- Data structure: Duplicate signatures combine both reads’ sequences, strands, and reference positions, with CIGAR parsing used to define positions accurately.The implementation splits signatures across arrays and hash tables keyed by sequence and strand combinations.
- Data structure: The custom data structure requires approximately 20 bytes per read pair, giving SAMBLASTER low memory requirements relative to other tools.Its hash tables store reference-position pairs as optimized 64-bit integers.