CIGAR String Encoding

The CIGAR string uses a defined set of single-character operations to describe an alignment. Each operation is preceded by a length integer, forming a compact run-length encoding.

• M (Match/Mismatch): An alignment match, which can represent a sequence match or a mismatch. The read base is aligned to the reference.
• I (Insertion): Bases present in the read but absent from the reference genome.
• D (Deletion): Bases present in the reference but absent from the read. This operation consumes reference bases but no read bases.
• N (Skipped Region): Represents an intron or other large gap in the read's alignment to the reference, typically used in spliced alignments.
• S (Soft Clipping): Read bases at the start or end that are not aligned but are retained in the Sequence Alignment Map (SAM) record's sequence field.
• H (Hard Clipping): Read bases that are not aligned and are also removed from the SAM record's sequence field.

A CIGAR string is parsed by iterating through pairs of numbers and operation characters. The example 100M1I50M1D25M describes a 176-base read with a complex alignment.

• 100M: The first 100 bases of the read align to the reference with matches or mismatches.
• 1I: The next base in the read is an insertion relative to the reference. It consumes 1 read base and 0 reference bases.
• 50M: The subsequent 50 read bases align to the reference.
• 1D: A single base is deleted from the reference. It consumes 1 reference base and 0 read bases.
• 25M: The final 25 read bases align to the reference.

This compact representation allows tools to rapidly reconstruct the exact gapped alignment without storing the full pairwise alignment matrix.

A critical property of the CIGAR string is the distinction between read-consuming and reference-consuming operations. This distinction is essential for calculating alignment length and mapping coordinates.

• Read-Consuming Operations: M, I, S, and =/X (sequence match/mismatch) advance the position in the read sequence.
• Reference-Consuming Operations: M, D, N, and =/X advance the position on the reference genome.
• Non-Consuming Operations: H and P (padding) consume neither read nor reference.

The sum of lengths for read-consuming operations must equal the length of the SEQ field in the SAM record. The sum of reference-consuming operations defines the total span of the reference covered by the alignment, which is used to calculate the mapping quality and identify structural variants.

The original CIGAR specification has been extended to support the complex alignment features of long-read sequencing technologies like PacBio and Oxford Nanopore.

• = (Sequence Match): An exact base-for-base match, providing finer resolution than the generic M operator.
• X (Sequence Mismatch): An explicit base substitution, allowing variant callers to immediately identify single nucleotide polymorphisms from the CIGAR string alone.
• P (Padding): A silent deletion from a padded reference, used in multiple sequence alignments where the reference has been artificially extended.

Using = and X instead of M transforms the CIGAR string from a coarse alignment summary into a precise edit transcript, enabling rapid variant identification without re-examining the base-level alignments.

Variant callers like DeepVariant and GATK HaplotypeCaller rely heavily on the CIGAR string to construct pileup images and perform local reassembly.

• Pileup Generation: The CIGAR string dictates which read bases are stacked at each reference position. Insertions (I) cause extra bases to be placed between reference columns, while deletions (D) create gaps.
• Indel Realignment: Regions with clusters of I and D operations indicate potential misalignments around true indels. Tools perform local reassembly to resolve these regions.
• Strand Bias Detection: Comparing the CIGAR operations from forward-strand reads versus reverse-strand reads at a locus can reveal systematic sequencing artifacts.

A malformed CIGAR string that violates consumption rules will cause pipeline failures, making strict validation a critical preprocessing step.

The CIGAR string is often paired with the optional MD tag in SAM/BAM files to provide a complete picture of the alignment.

• CIGAR String: Encodes the operations (match, insert, delete) but does not, by default, specify the reference bases that were mismatched.
• MD Tag: A string that encodes the reference bases for mismatches and deletions. For example, MD:Z:10A5^C3 indicates 10 matches, a mismatch to 'A', 5 matches, a deletion of 'C', and 3 matches.

Together, the CIGAR string and MD tag allow a tool to perfectly reconstruct the reference sequence at the aligned locus without accessing the full reference genome file, which is critical for the high-speed operation of variant callers and alignment viewers.

The process of discarding reads with a low probability of correct alignment before variant calling. The MAPQ field in SAM/BAM files is a Phred-scaled probability that the read is misaligned. A MAPQ of 30 corresponds to a 1/1000 chance of error. Filtering on MAPQ removes reads with ambiguous CIGAR strings that would otherwise introduce false positive variant calls due to mismapping in repetitive regions.

A method of encoding the vertical stack of aligned reads at a candidate locus into a multi-channel tensor for direct input into a convolutional neural network. Unlike traditional variant callers that parse CIGAR strings to count alleles, neural pileup representations preserve the spatial arrangement of reads, base qualities, strand information, and CIGAR-derived alignment features as distinct image channels, enabling the model to learn complex error patterns.

A targeted computational method that performs de novo assembly of reads mapping to a specific region to resolve complex variants. When CIGAR strings indicate multiple overlapping indels or soft-clipped reads that suggest a structural rearrangement, local reassembly constructs a consensus sequence from the reads themselves rather than forcing alignment to the reference. This resolves haplotypes that cannot be represented by a simple CIGAR string.

A systematic sequencing error where the true number of consecutive identical bases is miscounted, leading to false insertion or deletion calls. In CIGAR strings, this manifests as incorrect indel operation lengths in homopolymer runs (e.g., AAAAAA). The error arises from signal phasing issues in sequencing-by-synthesis platforms and is a major source of false positive indels that variant callers must model and correct.