BAM File

BAM files achieve 60-80% size reduction compared to their SAM text equivalents through BGZF block compression. This block-level gzip compression allows each block to be decompressed independently, enabling random access without decompressing the entire file. The compression preserves all original alignment data, including read sequences, quality scores, and CIGAR strings, while dramatically reducing storage costs for large-scale sequencing projects.

A companion BAI index file enables high-speed random access to any genomic region without scanning the entire BAM. The index stores binned reference coordinates mapping genomic positions to virtual file offsets. Tools like samtools view chr1:1000000-2000000 use this index to jump directly to reads overlapping the requested locus, making BAM the standard format for interactive genome browsers and targeted variant inspection.

Each aligned read carries a bitwise FLAG field encoding 12 boolean properties in a single integer:

• 0x1: Read is paired
• 0x2: Proper pair alignment
• 0x4: Read unmapped
• 0x8: Mate unmapped
• 0x40: First in pair
• 0x80: Second in pair
• 0x100: Secondary alignment
• 0x200: Fails quality checks
• 0x400: PCR duplicate
• 0x800: Supplementary alignment

Bitwise operations enable rapid filtering of reads by alignment status, strand, or duplicate state.

BAM supports extensible per-read metadata through optional auxiliary tags stored as key:type:value triplets. Common tags include:

• NM: Edit distance to reference
• MD: Mismatch position string
• AS: Alignment score
• XS: Suboptimal alignment score
• RG: Read group identifier
• BC: Barcode sequence

These tags carry quality metrics, alignment scores, and sample multiplexing information critical for downstream variant calling and deduplication.

BAM's read group header structure organizes reads by sample, library, platform unit, and sequencing run. Each read group contains metadata fields:

• ID: Unique group identifier
• SM: Sample name
• LB: Library preparation
• PL: Platform/technology
• PU: Platform unit (flowcell lane)

This structure enables multi-sample joint analysis, batch effect correction, and traceability from variant calls back to original sequencing runs.

BAM files are typically coordinate-sorted by reference sequence name and position, enabling efficient streaming algorithms for:

• Pileup generation for variant callers like DeepVariant
• Coverage depth calculation across contiguous regions
• Merging multiple BAMs from different libraries
• Duplicate marking using positional clustering

Coordinate sorting is a prerequisite for most downstream analysis tools and is enforced by the SAM specification for indexed files.

A compact, idiosyncratic string stored within each alignment record that summarizes the sequence of operations required to align a read to the reference. Each operation is represented by a length and a single-letter code:

• M: Alignment match (can be a sequence match or mismatch)
• I: Insertion in the read relative to the reference
• D: Deletion from the reference
• N: Skipped region (intron)
• S: Soft-clipped bases (not part of the alignment)
• H: Hard-clipped bases (removed from the read) The CIGAR string 150M indicates a perfect 150-base alignment, while 20S130M indicates 20 soft-clipped bases followed by 130 matching bases.

A Phred-scaled probability stored in every BAM record indicating the confidence that the read is correctly placed at its reported genomic position. A MAPQ of 30 means a 1 in 1,000 chance of incorrect mapping; a MAPQ of 0 indicates a multi-mapping read with no unique best placement. This field is critical for variant calling pipelines, where reads with low MAPQ are typically filtered out to prevent false positive variant calls caused by mismapped reads in repetitive regions.

A 12-bit integer stored in each BAM record that encodes multiple boolean properties of the alignment in a single value. Each bit represents a specific flag:

• 0x1: Read is paired
• 0x2: Read mapped in proper pair
• 0x4: Read unmapped
• 0x10: Read on reverse strand
• 0x40: First in pair
• 0x80: Second in pair
• 0x100: Secondary alignment
• 0x400: Read fails quality checks Tools like samtools flagstat decode these bitmasks to generate alignment summary statistics.