Mapping Artifacts in Difficult Genomic Regions

Why Read Placement Matters

A sequencer reports short reads, not their genomic addresses. An alignment algorithm places each read against a reference genome, and a variant caller uses the resulting alignments to infer which alleles are present at a locus. That inference depends on a crucial condition: the reads supporting a call must come from the genomic position being analyzed. In difficult regions, that condition can fail even when the underlying base calls are accurate.

Paralogs, repeated sequence, and differences between a sample’s genome and the reference can make distinct genomic locations look similar to the aligner. Reads may then be misplaced, split across alignments, or excluded. The result can be a false rare-variant call, a missed real variant, or an uncertain genotype. Read-level mapping quality and strand-bias diagnostics help identify these problems, but neither diagnostic by itself proves that a call is wrong or right.

How Difficult Regions Produce Erroneous Calls

When a read matches several locations similarly well, an aligner may not be able to determine its true origin. If a read from a paralog is assigned to the locus of interest, a base that differs between the two copies can appear to be an alternate allele at the target. If many reads from the same other copy are placed there, the apparent allele may have consistent support and still be artifactual. The central issue is not simply whether the base was read accurately; it is whether the read came from the locus to which it was assigned.

Repetitive sequence creates a related ambiguity. A short read from one repeat copy may match several copies, especially if the copies differ only outside the portion covered by that read. Reads that cannot be placed uniquely may receive low mapping quality, be assigned arbitrarily, or be omitted from an analysis. This can leave misleading support for a call or reduce the evidence available to detect a real variant.

A reference-genome mismatch is a different but sometimes overlapping problem. The reference may not represent a sequence carried by a sample, including a longer haplotype or an alternate structural arrangement. Reads from that sequence may contain many mismatches relative to the reference, align poorly, or align to a similar but incorrect location. Conversely, alignment can favor reads that match the reference more closely. This reference bias can make an alternate allele harder to observe, so mapping artifacts can cause false positives and false negatives.

A variant’s rarity does not protect it from these mechanisms. Indeed, an error affecting only a few samples can resemble a rare allele. A plausible genotype likelihood based on observed read bases is not enough if the reads may have come from elsewhere; genotype evidence depends on the reliability of read placement as well as on base calls.

Mapping Quality: Evidence About Placement

Mapping quality, often abbreviated MAPQ, is a read-level measure of confidence in an alignment. It is commonly expressed on a Phred scale: larger values indicate a lower estimated probability that the read was placed incorrectly. In the idealized scale, a mapping quality \(Q\) corresponds to an estimated placement-error probability \(e\) through the following relationship.

For example, \(Q=30\) corresponds to \(e=10^{-3}=0.001\), or a nominal one-in-one-thousand placement-error probability. This is a way to interpret the scale, not a guarantee that every program’s reported MAPQ is perfectly calibrated. Aligners differ in how they calculate and cap values, and some difficult regions are not adequately represented in the reference used by the aligner.

Mapping quality is not the same as base quality. Base quality concerns the probability that a sequenced nucleotide is wrong; mapping quality concerns the probability that the read is assigned to the wrong genomic location. A read can have high base quality and low mapping quality: its sequence may be read accurately, but several genomic locations may match it. Nor is MAPQ a probability that a genotype is correct. Genotype inference also depends on the pattern of reads, sequencing errors, and the assumptions of the calling procedure.

Strand-Bias Diagnostics

Sequencing reads are observed in forward and reverse orientations relative to the reference. For a genuine variant with adequate, technically sound read support, alternate alleles are generally expected to be observed in both orientations. A strong imbalance can point to an artifact associated with read orientation, library preparation, sequence context, or alignment. It is a diagnostic pattern, not a universal rule: chance imbalance can occur, especially with few reads, and some genuine calls will have uneven strand counts.

A simple diagnostic compares the numbers of reference- and alternate-supporting reads on each strand. One can also compare the fraction of reads carrying the alternate allele separately by strand. Formal strand-bias metrics may use a contingency-table test or other transformations; their numerical values and thresholds depend on the method. The underlying question is whether allele support is unusually concentrated in one orientation relative to the rest of the read evidence.

Putting the Diagnostics Together

Mapping quality and strand bias address related but distinct concerns. Low MAPQ suggests that a read has plausible alternative placements. Strand bias suggests that support for an allele depends strongly on read orientation. Either pattern can occur without the other. For instance, a read may map confidently but have orientation-specific evidence, while reads from repeated sequence may have low MAPQ on both strands.

A useful review asks where the reads align and what the alignment looks like, not only how many reads support each allele. Nearby mismatches, soft-clipped read ends, inconsistent alignment positions, or reads that appear to match a homologous region can help explain why support is suspicious. Paired reads and longer reads may provide placement clues if their other ends or full sequences are more specific. Comparing alignments against an appropriate alternate representation of the region can also be informative.

These checks do not substitute for a complete assessment of a variant call. They indicate whether the evidence may be compromised by placement. A call with strong evidence from uniquely placed reads on both strands is more reassuring than one supported by ambiguous or orientation-skewed reads, but no single diagnostic establishes correctness. A high MAPQ can be misleading if the aligner’s reference omits the true alternative sequence, and balanced strand counts do not rule out misalignment.

A Short Justification for the Mapping-Quality Scale

The Phred transformation makes small error probabilities easier to represent as positive integer scores. Starting from \(Q=-10\log_{10}(e)\), divide both sides by \(-10\) to obtain \(\log_{10}(e)=-Q/10\). Raising 10 to both sides gives \(e=10^{-Q/10}\). Each increase of 10 in MAPQ therefore reduces the nominal error probability by a factor of 10: \(Q=20\) corresponds to 0.01, \(Q=30\) to 0.001, and \(Q=40\) to 0.0001. This arithmetic explains the scale; it does not validate the calibration of a particular aligner in a particular genomic region.

Common Mistakes and Practical Interpretation

• Equating base quality with mapping quality. A high-quality base can belong to a read assigned to the wrong locus. Check both the nucleotide evidence and read placement.
• Treating MAPQ as the probability the genotype is wrong. MAPQ concerns read placement, not the posterior probability of a genotype. Genotype confidence requires additional evidence and a model.
• Assuming high MAPQ settles the question. An aligner can assign high confidence relative to an incomplete or unsuitable reference. Consider whether the region’s paralogs, repeats, or alternate sequences are represented.
• Calling a variant false whenever strand counts are unequal. Unequal counts can arise by chance, particularly with sparse read support. Interpret the imbalance in context and inspect alignments.
• Assuming support on both strands rules out mapping artifacts. Misplaced reads can occur in both orientations. Strand balance is one diagnostic, not a guarantee of correct placement.
• Interpreting a rare call as biologically plausible before checking its location. A small number of erroneous reads can mimic rarity. First ask whether the supporting reads plausibly originate from the locus.