Sequence formats are not interchangeable. Each captures a different stage of analysis, and converting between them can discard information.
The useful question is not “which format is best?” but “what evidence must survive this handoff?” A raw read, an accepted consensus, and a list of differences from a reference answer different questions. Giving all three the same sample name does not make them equivalent.
The formats at a glance
| Format | Primary unit | Preserves | Does not by itself preserve | Typical mtDNA role |
|---|---|---|---|---|
| AB1 | one capillary trace | called bases, quality values, fluorescence channels, ABIF metadata | paired-read reconciliation or an accepted consensus | Sanger evidence and trace review |
| FASTQ | one or many reads | sequence plus encoded per-base qualities | capillary trace shape, alignment, reference coordinates | raw or processed reads, sometimes a quality-bearing consensus |
| FASTA | one or many sequences | identifier, description, sequence | per-base quality, read depth, variant filters | consensus exchange, alignment, classification |
| VCF | sites differing or assessed against a reference | coordinates, reference/alternate alleles, filters, INFO and optional sample fields | the complete sample sequence unless reconstructed under explicit rules | variant exchange and annotation |
AB1: retain the observable trace
AB1 is an Applied Biosystems binary container. The exact tags vary by instrument and software, but files commonly include base calls, Phred-like quality values, peak locations, analyzed dye-channel arrays, and run metadata. A FASTA exported from the same file keeps the letters but loses the evidence needed to inspect peak shape, baseline, and secondary signal.
Use AB1 as the source artifact for Sanger data. Name it with a stable sample and primer/read identifier, calculate a checksum, and record strand and amplicon separately rather than trying to encode the whole experiment in a filename. Never discard the AB1 after generating a consensus FASTA.
FASTQ: sequence plus quality, with a major ambiguity
Each FASTQ record has an identifier, sequence, separator, and encoded quality string. Before using it, determine what one record represents. It could be a single instrument read, a trimmed read, an assembled contig exported with consensus qualities, or an entire raw high-throughput run containing millions of reads. Those inputs require very different processing.
Confirm the quality encoding, although modern data normally use Sanger/Illumina 1.8+ Phred+33. Confirm whether reads are paired and whether pairing is encoded in names or separate files. Also preserve the preprocessing history: adapter removal, quality trimming, read merging, mapping, duplicate handling, and consensus generation. FASTQ quality values do not store these operations.
Raw short-read mtDNA data ordinarily require a validated external workflow for read QC, mapping or assembly, NUMT-aware handling where relevant, depth assessment, contamination checks, variant calling, and consensus generation. Treating each raw read as an independent biological sample is a category error.
FASTA: simple, useful, and evidence-poor
FASTA is appropriate when the deliverable is a sequence: an accepted consensus, reference, primer, or alignment. The definition line is free text, so establish an identifier policy. Use a stable, non-identifying sample code as the first token and put structured metadata in a manifest rather than relying on prose headers that downstream tools may truncate.
Record whether a sequence is complete or partial, its orientation, molecule topology, coordinate span, reference used for numbering, and ambiguity alphabet. Uppercase/lowercase and gap symbols do not have universal semantics. N means an unresolved base in many workflows; IUPAC codes such as R or Y represent defined sets of possible bases, but they do not communicate read depth or mixture proportion.
A FASTA derived from AB1 or FASTQ should link back to those sources and to the rule that produced each consensus position. Without that relationship, a polished sequence can conceal low quality and manual edits.
VCF: a reference-relative claim
A VCF row is not merely “a mutation.” CHROM, POS, REF, and ALT only make sense against the exact reference sequence and coordinate convention used by the caller. The header should identify the reference and define INFO, FILTER, and FORMAT fields. Sample columns can carry genotype, depth, allele depth, genotype quality, and other caller-specific values, but none is guaranteed to exist.
For mtDNA, inspect these decisions before comparison:
- Was rCRS, RSRS, or another sequence used, and is its accession/version recorded?
- Was the mitochondrial molecule treated as circular near the coordinate origin?
- Were indels left-aligned and normalized, and how were homopolymers represented?
- Are homoplasmic and heteroplasmic calls represented as genotypes, allele fractions, or custom fields?
- Which filters were applied, and do missing sites mean reference, no coverage, or not assessed?
- Were multi-allelic sites split, and were strand/position annotations normalized?
Never infer a full consensus by applying only PASS variants unless callable intervals and no-call rules are also known. In a sparse VCF, absence of a row is not proof of the reference allele.
Choose the handoff deliberately
- Preserve the source. Keep AB1 or raw FASTQ unchanged with checksum and acquisition metadata.
- Create analysis derivatives. Trim, align, assemble, or call variants in a versioned workflow. Do not overwrite the source.
- Exchange the right object. Use FASTA for an explicit consensus, VCF for reference-relative calls with headers, and FASTQ when qualities are needed.
- Ship a manifest. Map file/record identifiers to sample, source artifact, processing run, reference accession/version, coordinate span, software versions, parameters, and checksums.
- Validate after conversion. Count records and bases, compare identifiers, validate alphabets, and spot-check consequential sites against the upstream evidence.
Common lossy conversions include AB1 to FASTA (trace and quality lost), FASTQ to FASTA (quality lost), BAM/CRAM to VCF (nonvariant evidence and most read context lost), and VCF to FASTA (result depends on coverage, ploidy, filtering, ambiguity, and indel rules). Loss is acceptable only when it is intentional and the derivative remains linked to its source.
Where GeneFlow fits
GeneFlow’s Data Vault stores uploaded AB1, FASTA, FASTQ, and VCF artifacts with project, uploader, kind, size, checksum, object-storage key, processing status, and parse errors. Parsing links resulting samples to the source file. FASTA records populate a sample consensus sequence. AB1 records populate a trace model. VCF sample columns create source-linked samples and are handed to HaploGrep3 as reference-relative input.
The current FASTQ parser is intentionally not a raw NGS pipeline: it accepts at most 500 records, treats each record as a sample sequence with qualities, and rejects larger read sets. Researchers must assemble/process raw NGS data externally and upload an appropriate consensus FASTA/FASTQ or reviewed VCF. GeneFlow does not normalize VCFs, reconstruct full consensus from variants, infer missing-site callability, or prove that a declared reference is correct. It stores and routes evidence; format validation and scientifically valid conversion rules remain part of the research workflow.