Mapping Quality Filtering

The Mapping Quality (MAPQ) score is a Phred-scaled probability that the read is misaligned. A score of 30 implies a 1 in 1,000 chance of incorrect alignment, while a score of 0 indicates a multi-mapping read with no unique best hit. Filtering on a threshold (typically MAPQ ≥ 20 or 30) is the primary mechanism for removing ambiguous reads from repetitive or paralogous regions.

Beyond the aggregate MAPQ score, filtering often involves analyzing the edit distance or mismatch density of an alignment. A read with an unexpectedly high number of mismatches or gaps relative to the reference may represent a pseudogene or a structural variant rather than a true single-nucleotide variant. Hard-clipping or discarding reads with excessive soft-clipped bases prevents false variant calls near insertion/deletion breakpoints.

Short reads originating from low-complexity regions or segmental duplications often align equally well to multiple genomic loci. Aligners typically assign a MAPQ of 0 to these reads. Filtering them out is essential to avoid spurious heterozygous calls caused by aggregating signal from paralogous sequences. Advanced pipelines may also use a k-mer uniqueness track to pre-emptively mask problematic regions.

A single sequenced fragment can produce multiple alignment records: a primary linear alignment and supplementary chimeric or split-read alignments. For standard variant calling, supplementary alignments must be flagged and filtered out (using SAM flag 2048) to prevent double-counting evidence. However, these reads are retained for structural variant detection to identify breakpoints.

Standard MAPQ applies to the entire read, but local alignment artifacts near indels can cause systematic base-level errors. Base Alignment Quality (BAQ) is a computational adjustment that lowers the quality score of individual bases hidden within a gapped alignment. This prevents false single-nucleotide variant calls caused by the misplacement of insertions or deletions by the aligner.

Aggressive mapping quality filtering is the most effective single step for controlling the False Discovery Rate (FDR) in variant calling. By removing reads with MAPQ < 20, pipelines can eliminate the majority of false positives arising from mismapped reads in segmental duplications. However, overly strict filtering can cause allelic dropout in highly polymorphic regions, reducing sensitivity for true variants.