A Provenance Layer for Ancient DNA

Executive summary

Ancient DNA has become one of the most productive fields in the life and historical sciences. Genome-wide data from more than 10,000 ancient individuals had been published by the end of 2022, and the number continues to grow. Increasingly, the field’s conclusions come not from single studies but from the reuse of compiled datasets by researchers who never handled the original remains.

That success creates a problem. Every ancient genome is the product of a long evidence chain — from site and archaeological context, through specimen, sample, extract, library and sequencing run, to processed data, community datasets, publications and reanalyses. Each link is recorded somewhere, in a different system, under different identifiers. The relationships between those records — which library came from which sample, which archive accession corresponds to which individual in a paper, which version of a record an analysis used — are often recorded nowhere reliable at all.

The evidence that this matters is now direct. A 2024 audit of 42 ancient genomics studies published in leading journals found that more than a third had archive accessions that did not match the paper’s description of its samples; fewer than half archived raw sequencing reads; and only one provided radiocarbon laboratory codes in the archive. The community resources that compile ancient genomes are careful and valuable, but they are explicit that identifiers change between releases and that corrections are made going forward.

This white paper proposes a provenance layer for ancient DNA: an open, versioned record of the events and relationships that connect existing records across systems. It does not store or replace sequence data, archaeological records or curated datasets. It records how they relate, how they changed, and on what basis each link is asserted.

The paper sets out:

• the problem, and the evidence for it (sections 1–2);
• why existing infrastructure, excellent as it is, does not solve it (section 3);
• a provenance model built on the W3C PROV standard, with events, versioning and graded link confidence (section 4);
• a proposed minimum provenance record for every ancient genome (section 5);
• the questions a provenance layer makes answerable (section 6);
• a pilot that reconstructs evidence chains from public data and measures where they break (section 7);
• governance, ethics and sustainability, including the rights of descendant communities (section 8).

We are publishing it as a draft for comment and are seeking partners among laboratories, data curators, museums and funders to test the pilot.

1. The problem: from sequence to evidence

1.1 An ancient genome is a chain, not an object

Consider a single ancient individual. A tooth is excavated from a burial and given a context number on site. The remains enter a museum collection under an accession number. A laboratory samples the tooth, extracts DNA and prepares one or more sequencing libraries, each with its own code. Sequencing produces raw reads, which are trimmed, aligned to a reference genome, filtered and assessed for damage and contamination. The individual appears in a publication under a label chosen by the authors. The data are deposited in a public archive, under sample, experiment and run accessions. Later, the individual is incorporated into a community dataset under another identifier, and from there into dozens of further analyses.

What is commonly called “an ancient genome” is the product of that chain. Change something at one link — a sample reassignment, a revised date, a different filter — and conclusions downstream may change with it.

1.2 The problem of distance

Large-scale ancient DNA research depends on reuse. A researcher analysing population history across a continent may use thousands of individuals from hundreds of studies, compiled by others. That is what makes the field powerful. It also means that the people drawing conclusions are increasingly distant from the specimens, laboratories and decisions that produced the data.

When researchers can no longer know the history of every observation personally, the infrastructure has to preserve it for them.

1.3 Why ancient material raises the stakes

Ancient biological material is finite, often irreplaceable and frequently destroyed in sampling. DNA is degraded and contamination can materially change results. Dating and archaeological interpretation are revised as methods improve. And the individuals concerned are people whose descendants and communities may have legitimate interests in how their remains are used. A lost link in the chain cannot simply be re-created by taking another sample.

2. What the evidence shows

The problem is no longer anecdotal. The field has begun to audit itself, and the findings are consistent.

Archiving practice. In 2024 Anders Bergström examined the archived data from 42 ancient genomics studies published in Nature, Science and Cell between 2021 and 2023.¹

Linking samples to accessions. The team behind AncientMetagenomeDir, a community-curated index of published ancient microbial and environmental samples, reported “errors in some data accessions uploaded to public repositories, such as multiple sample codes assigned to different libraries of the same sample, and insufficient metadata to link accessions to specific samples reported in a study”.²

Identifiers over time. The Allen Ancient DNA Resource (AADR), the largest compiled resource of ancient human genotype data, states that its identifiers “may change from data release to data release” and that when two identifiers are found to refer to the same individual, “they are merged into a single Master ID”. It relies on ongoing curation and user feedback to identify errors, “which we then seek to correct in subsequent releases”.³

Decay. Outside ancient DNA, a widely cited study by Vines and colleagues found that where authors reported the status of their data, the odds of a dataset still existing fell by 17% for every year since publication, and the odds of finding a working email address for the authors fell by 7% a year.⁴ Context held only in supplementary files, or only by the authors, is not durable.

Taken together, these findings describe five recurring breaks, discussed in detail in the Origins article Five Ways an Ancient Genome Gets Lost:

• Identity — identifiers multiply across systems and change over time.
• Link — publications and archives disagree about which data belong to which individual.
• Raw evidence — unaligned reads and failed samples are not archived.
• Context — place, time and dating information stay in supplementary files.
• Versions — corrections propagate forward but not to copies already made.

None of these is primarily a failure to collect data. Each is a failure to record a relationship.

3. Why existing infrastructure is not enough

The ancient DNA community has built excellent infrastructure. A provenance layer should build on it, not compete with it.

Each resource is authoritative within its own domain.^3, 2, 5 The gap lies between them. No existing system is responsible for recording that museum accession A, laboratory sample B, library C, archive run D, publication label E and dataset identifier F all refer to the same individual; that F was merged with G in a later release; that a particular analysis used the version before the merge; and on what evidence each of those statements rests.

That is the job of a provenance layer.

4. The Ancient BioChain model

4.1 Design principles

Connect, don’t replace. Existing identifiers remain authoritative. An ENA accession stays an ENA accession; a DOI stays a DOI; a museum number stays meaningful in its museum. The provenance layer records relationships between them and creates its own identifiers only for the events and links it records.

Events, not just records. Provenance describes what happened. The model records sampling, extraction, library preparation, sequencing, processing, curation, publication and reanalysis as events connecting one thing to the next.

Nothing is overwritten. Corrections are recorded as new events that supersede earlier ones, never as silent edits. Every earlier state remains retrievable, and each record carries a content fingerprint — a cryptographic hash of the record — so that later users can confirm it has not changed since it was cited.

Every link carries its evidence. A relationship asserted by the original authors is different from one inferred by a curator, which is different again from one matched automatically. The model records which is which.

Open by default, protected where necessary. Provenance records should be openly accessible, with protections for sensitive information such as precise burial locations and for community-governed data (section 8).

4.2 Standards alignment

The model follows the W3C PROV data model, a web standard for provenance,⁶ which describes the world in terms of:

• entities — things such as specimens, samples, libraries, files, dataset records and publications;
• activities — events such as sampling, sequencing, processing or curation, which use some entities and generate others;
• agents — the people, laboratories, institutions and software responsible for activities.

Using PROV means records can be exchanged with other provenance-aware systems and queried with standard tools. The model is also designed to support the FAIR principles — making data findable, accessible, interoperable and reusable⁷ — by adding the relationships that FAIR metadata alone does not capture.

4.3 Core entities and events

Agents are identified with existing persistent identifiers where possible — ORCID for people and ROR for institutions.

4.4 Link confidence

Every relationship in the graph records how it is known:

• Asserted — stated explicitly by the original data producer, for example in archive metadata.
• Documented — stated in a publication or supplementary table but not in structured metadata.
• Curated — established by a curator from multiple sources, with the reasoning recorded.
• Inferred — matched automatically, for example by identifier similarity or sequence comparison, with the method recorded.
• Disputed — conflicting sources exist; both are recorded.

This makes gaps visible rather than hiding them. A chain made entirely of asserted links is very different from one held together by inference, and users can filter on that distinction.

4.5 Versions and corrections

When a record changes — a date recalibrated, two individuals merged, a sample reassigned, a contamination flag added — the layer records a new event linking the new state to the old one, with the reason and the source. Analyses record which version they used. It then becomes possible to ask which published results depend on a version that has since been superseded, and to notify their authors.

5. A minimum provenance record

Much of the value of a provenance layer depends on the information producers record in the first place. We propose the following minimum provenance record for every published ancient genome, organised by the five breaks. It is deliberately modest: almost all of it already exists somewhere in a typical study.

The Ancient DNA Provenance Checklist, published alongside this paper, turns this record into a practical checklist for data producers and users. The record builds directly on the archiving recommendations in Bergström (2024)¹ and on metadata practices already used by AncientMetagenomeDir² and Poseidon.⁵

6. What becomes possible

A provenance layer changes the questions that can be asked of the scientific record. Examples include:

• Trace: show everything that happened between specimen X and result Y.
• Resolve: which identifiers, across all systems, refer to this individual?
• Audit: which individuals in this dataset lack archived raw reads, a radiocarbon laboratory code or a sampling permission?
• Assess: for this analysis, what share of the evidence chain is asserted, documented, curated or inferred?
• Propagate: which published analyses used a version of an individual that has since been merged, re-dated or flagged as contaminated?
• Discover: which archived human datasets retain unaligned reads that could be screened for ancient pathogens?
• Credit: which laboratories, museums and curators contributed to the evidence behind a given result?

For ancient pathogen research, where a pathogen genome may be recovered from reads originally sequenced for human population analysis — as in the recovery of Bronze Age Yersinia pestis genomes from existing shotgun data⁹ — and then combined with modern isolates, the ability to trace a phylogenetic conclusion back to the archaeological specimen is particularly valuable. The same logic connects ancient DNA to modern genomic surveillance: in both, the question is whether we can show where the evidence came from.

7. Pilot: reconstructing chains from public data

7.1 Aim

The first phase will test how much of the evidence chain behind published ancient genomes can be reconstructed from public information alone, and where it breaks.

7.2 Scope

A defined set of published individuals, selected to include a range of publication dates, laboratories and data types (shotgun and capture; human and pathogen), drawn from public compiled datasets and cross-linked to sequence archives, publications and, where available, collection records.

7.3 Method

• Extract identifiers and metadata from compiled datasets, sequence archives and publications.
• Build the provenance graph, recording each link with its confidence class.
• Record every break by type — identity, link, raw evidence, context, versions — and the effort required to resolve it.
• Test the example queries in section 6.
• Invite data producers to confirm or correct a sample of inferred links.

7.4 Measures

• Share of individuals with a complete chain from specimen to archived raw reads.
• Distribution of link confidence classes.
• Frequency of each break type, by publication year and data type.
• Time and effort needed to resolve a break manually.
• Accuracy of automated inference against producer-confirmed links.

7.5 Outputs

An open provenance graph for the pilot set; a report on where evidence chains hold and where they break; a worked, traceable case study of a single individual; and a revised minimum provenance record informed by what the pilot finds.

7.6 Phases

8. Governance, ethics and sustainability

8.1 Human remains and descendant communities

Ancient DNA research concerns people, and increasingly the field recognises obligations to the communities connected to them. In 2021, an international group of researchers set out globally applicable ethics guidelines for DNA research on human remains, covering compliance with regulations, detailed planning, minimising damage to remains, data availability after publication, and engagement with stakeholders.¹⁰ The CARE Principles for Indigenous Data Governance — Collective benefit, Authority to control, Responsibility and Ethics — complement the FAIR principles by addressing who governs data about Indigenous peoples.¹¹

A provenance layer can support these obligations. Recording sampling permissions, consultation and data-use conditions as part of the chain makes them visible to every later user, rather than confining them to the original study.

8.2 Sensitive information

Precise burial locations can put sites at risk of looting or disturbance, and some communities may not wish certain information to be public. The model allows location and other sensitive fields to be recorded as protected references, accessible under conditions set by the responsible institution or community, while the rest of the chain remains open.

8.3 Openness

Provenance records will be openly licensed and exportable in standard formats. The model and software will be developed openly so that the community can inspect, adopt and extend them.

8.4 Sustainability

A provenance layer is infrastructure and needs long-term stewardship. We expect its governance to involve the communities that produce and curate ancient DNA data, and its funding to combine research grants in the early phases with institutional partnerships and services for data producers in later ones. It sits within the wider architecture of The BioChain, which applies the same principle — connecting evidence rather than replacing the systems that hold it — to modern biological data.

9. Who benefits

• Laboratories and authors: clearer credit, easier compliance with journal and funder data requirements, and fewer queries about which file belongs to which sample.
• Data curators: a record of identifier changes and corrections that users can follow, reducing repeated curation.
• Researchers reusing data: confidence about exactly what they are analysing, and alerts when inputs change.
• Museums and collections: a record of what has been done with sampled material and what it produced.
• Journals and funders: a practical standard against which data availability can be checked.
• Pathogen and surveillance researchers: traceable links from phylogenetic conclusions back to specimens.
• Descendant communities: visibility of permissions and conditions throughout the chain.

10. Limitations and what would prove this wrong

A provenance layer cannot recover information that was never recorded. Where a laboratory did not keep the relationship between a library and a sample, the layer can only record that the link is inferred or missing.

It also depends on participation. Reconstruction from public data can show where chains break; closing those gaps requires data producers and curators to record relationships at the source.

The case for this work would be weakened if the pilot found that most evidence chains can in fact be reconstructed reliably and cheaply from existing sources; if the breaks identified proved to have little effect on scientific conclusions; or if community resources adopted cross-system provenance recording on their own, making a separate layer unnecessary. Each of those would be a good outcome for the field, and the pilot is designed to find out.

11. Next steps and invitation

We are publishing this white paper as a draft for comment. We welcome feedback on the model and the minimum provenance record, and we are seeking:

• laboratories willing to confirm or correct reconstructed chains for their published data;
• curators of compiled datasets and community indexes interested in exchanging provenance records;
• museums and collections interested in linking sampling events to derived data;
• funders and partners for the pilot.

The sequence is data. The chain that produced it is evidence.

References

1. Bergström, A. (2024) ‘Improving data archiving practices in ancient genomics’, Scientific Data, 11, 754. https://doi.org/10.1038/s41597-024-03563-y
2. Fellows Yates, J.A., Andrades Valtueña, A., Vågene, Å.J., et al. (2021) ‘Community-curated and standardised metadata of published ancient metagenomic samples with AncientMetagenomeDir’, Scientific Data, 8, 31. https://doi.org/10.1038/s41597-021-00816-y
3. Mallick, S., Micco, A., Mah, M., et al. (2024) ‘The Allen Ancient DNA Resource (AADR): a curated compendium of ancient human genomes’, Scientific Data, 11, 182. https://doi.org/10.1038/s41597-024-03031-7
4. Vines, T.H., Albert, A.Y.K., Andrew, R.L., et al. (2014) ‘The availability of research data declines rapidly with article age’, Current Biology, 24(1), pp. 94–97. https://doi.org/10.1016/j.cub.2013.11.014
5. Schmid, C., Ghalichi, A., Lamnidis, T.C., et al. (2024) ‘Poseidon – A framework for archaeogenetic human genotype data management’, eLife (reviewed preprint RP98317). https://elifesciences.org/reviewed-preprints/98317
6. Lebo, T., Sahoo, S. and McGuinness, D. (eds.) (2013) PROV-O: The PROV Ontology. W3C Recommendation, 30 April 2013. https://www.w3.org/TR/prov-o/
7. Wilkinson, M.D., Dumontier, M., Aalbersberg, I.J., et al. (2016) ‘The FAIR Guiding Principles for scientific data management and stewardship’, Scientific Data, 3, 160018. https://doi.org/10.1038/sdata.2016.18
8. Reimer, P.J., Austin, W.E.N., Bard, E., et al. (2020) ‘The IntCal20 Northern Hemisphere radiocarbon age calibration curve (0–55 cal kBP)’, Radiocarbon, 62(4), pp. 725–757. https://doi.org/10.1017/RDC.2020.41
9. Rasmussen, S., Allentoft, M.E., Nielsen, K., et al. (2015) ‘Early divergent strains of Yersinia pestis in Eurasia 5,000 years ago’, Cell, 163(3), pp. 571–582. https://doi.org/10.1016/j.cell.2015.10.009
10. Alpaslan-Roodenberg, S., Anthony, D., Babiker, H., et al. (2021) ‘Ethics of DNA research on human remains: five globally applicable guidelines’, Nature, 599, pp. 41–46. https://doi.org/10.1038/s41586-021-04008-x
11. Carroll, S.R., Garba, I., Figueroa-Rodríguez, O.L., et al. (2020) ‘The CARE Principles for Indigenous Data Governance’, Data Science Journal, 19, 43. https://doi.org/10.5334/dsj-2020-043