Building a Defensible mtDNA Consensus Sequence

A consensus sequence is a model of the bases supported by available reads. It is not simply the longest read or an automatic majority vote. The construction rules should reflect assay design and preserve uncertainty.

A defensible consensus answers three questions for every reported position: which evidence covered it, which rule selected the symbol, and where uncertainty was retained. It must not be improved by silently changing calls to match rCRS, RSRS, a haplogroup motif, or another expected result.

Define the deliverable first

Before combining reads, write down:

• whether the target is a complete mitogenome, control region, gene, or amplicon set;
• the reference accession and version used for coordinates, not merely “rCRS”;
• whether the output is a contiguous sequence or only covered intervals;
• the minimum evidence for a base call and for an ambiguity;
• rules for primer sequence, low-quality ends, internal gaps, indels, mixtures, and strand conflicts;
• whether the method is exploratory or governed by a validated protocol.

For high-throughput data, add minimum depth, mapping quality, base quality, strand balance, allele-fraction, contamination, and NUMT-handling rules. Those decisions need a validated external pipeline; a Sanger overlap procedure cannot simply be transferred to NGS data.

Prepare each read independently

1. Verify identity. Match file, sample, primer, strand, amplicon, and run records. Investigate duplicate or unexpected identifiers before sequence work.
2. Review source quality. For Sanger reads, inspect the AB1 trace and define an accepted interval. For NGS, review read, mapping, depth, and contamination QC from the processing workflow.
3. Remove non-biological sequence. Exclude primers and adapters using known sequences and documented matching rules. Primer-derived bases are not independent observations of the sample.
4. Orient consistently. Reverse-complement reverse reads so all records use the same mtDNA orientation. Keep the original orientation in metadata.
5. Retain a derivative. Save trimmed/oriented reads with links to immutable source files and the exact operation that generated them.

Do not start by aligning every raw base and then trimming whichever columns disagree. That makes the desired agreement determine the QC boundary.

Align homologous positions

Align accepted read intervals to one another and, when coordinates are needed, to the stated reference. Check that overlaps occur where the assay design predicts. Unexpected overlap length can reveal swapped primers, reverse-complement errors, or an off-target product.

Mitochondrial DNA is circular, but most software and references linearize it at a chosen origin. Amplicons crossing that origin may need to be rotated or split before alignment. Record the transformation so coordinates can be mapped back. Homopolymer-associated indels and length heteroplasmy deserve manual scrutiny because equally plausible gap placements can produce different variant descriptions while representing the same local sequence.

An aligner proposes homology; it does not decide whether a questionable base is genuine. Inspect alignment columns around every mismatch, indel, and low-quality interval against the underlying reads or traces.

Apply an explicit decision table

The exact thresholds belong to the assay, but the logic should be predetermined. A Sanger-oriented example is:

Distinguish three concepts that are often collapsed:

• Unknown base (N): evidence does not support a constrained set.
• IUPAC ambiguity: evidence supports a defined set of bases, such as R for A/G.
• Gap (-): an alignment character, not automatically a biological deletion and usually not part of an ungapped consensus FASTA.

For NGS consensus generation, a majority allele is not automatically sufficient. A 60:40 observation can reflect heteroplasmy, contamination, strand bias, mapping ambiguity, or systematic error. Report the depth and allele evidence separately from the single consensus symbol when the study question depends on within-sample variation.

Validate the assembled consensus

Run checks before downstream classification:

1. Confirm allowed characters, orientation, expected length, and covered coordinate range.
2. Map every non-reference call and ambiguity back to source evidence.
3. Check that primer-binding positions and excluded ends were not reintroduced.
4. Review unexpected clusters of differences, which can indicate register shifts, contamination, poor alignment, or sample exchange.
5. Compare haplogroup classification only as a diagnostic. Missing expected motifs or many private changes can trigger re-review, but must not dictate edits.
6. Have a second reviewer inspect consequential manual decisions when the protocol requires it.

Version the accepted consensus. If a trim boundary or call changes, create a new derivative and record why; do not replace the prior sequence without history.

Failure modes worth catching

Reference-biased correction. Replacing a weak call with the reference hides uncertainty and can manufacture a plausible haplotype.

False independence. Forward and reverse sequences from the same PCR help confirm sequencing calls, but they are not independent extraction or amplification replicates.

Coordinate drift. Mixing zero-based read indices, one-based reference positions, and gapped alignment columns creates off-by-one errors. Name each coordinate system.

Unqualified partial sequences. Downstream tools may classify a partial FASTA, but confidence and comparability depend on the observed range. Report what was not sequenced.

Consensus-only archiving. A FASTA without source links cannot support later review of qualities, traces, depth, filters, or edits.

Consensus reproducibility checklist

• Sample, source file identifiers, checksums, primer/strand, and replicate relationships
• Reference name, accession/version, topology handling, and coordinate conventions
• Accepted interval and trimming/primer-removal rule for every read
• Alignment software/version, command or parameters, and reviewed alignment artifact
• Base, ambiguity, indel, coverage, and conflict-resolution rules
• Per-position manual edits with original call, evidence, reason, reviewer, and date
• Final FASTA checksum, sequence length, covered range, and version
• Validation results and unresolved limitations

Where GeneFlow fits

GeneFlow keeps a sample linked to its uploaded source file and provides a field for the sample’s consensus sequence. AB1 uploads also create a one-to-one trace record containing base calls, qualities, channels, and metadata; FASTA and small FASTQ records populate the consensus field directly. Downstream haplogroup classification and project tree generation prefer the stored consensus and otherwise fall back to AB1 base calls.

That storage model supports evidence review, but GeneFlow does not currently assemble forward and reverse reads, trim primers, resolve overlaps, call a consensus from raw NGS reads, or adjudicate ambiguities. A sample has one primary source-file link and one AB1 trace record, so multi-read provenance and the full consensus decision log must be maintained in the research record or validated external workflow. Upload the reviewed consensus as a versioned derivative and retain all source evidence. GeneFlow can then connect that sequence to QC, HaploGrep3, tree, and report records without claiming that those outputs validated the consensus construction.

Further reading

• NC-IUB recommendations for ambiguity symbols in nucleic-acid sequences
• MITOMAP: human mitochondrial reference sequences
• Parson et al., DNA Commission guidelines for mitochondrial DNA typing