CoChat
Sign Up
Cited by 135DNA and Biological ComputingAdvanced Data Storage TechnologiesAdvanced biosensing and bioanalysis techniques

Probing the physical limits of reliable DNA data retrieval.

Open access
Organick, Lee, Chen, Yuan-Jyue, Dumas Ang, Siena, Lopez, Randolph, Liu, Xiaomeng, Strauss, Karin, Ceze, Luis·2020·Nature Communications
DOI 10.1038/s41467-020-14319-8PMC PMC6992699
Open this paper in CoChat — free accountOpen PDF

Abstract

Synthetic DNA is gaining momentum as a potential storage medium for archival data storage. In this process, digital information is translated into sequences of nucleotides and the resulting synthetic DNA strands are then stored for later retrieval. Here, we demonstrate reliable file recovery with PCR-based random access when as few as ten copies per sequence are stored, on average. This results in density of about 17 exabytes/gram, nearly two orders of magnitude greater than prior work has shown. We successfully retrieve the same data in a complex pool of over 10 10 unique sequences per microliter with no evidence that we have begun to approach complexity limits. Finally, we also investigate the effects of file size and sequencing coverage on successful file retrieval and look for systematic DNA strand drop out. These findings substantiate the robustness and high data density of the process examined here.

AI snapshot

Written from the paper’s full text.

TL;DR

PCR-based random access recovered three DNA-encoded files reliably when each sequence was present at an average of only 10 copies, yielding a density of about 17 EB g−1. The same data were retrieved from pools with over 10^10 unique sequences per microliter, with no evidence that the complexity limit had been reached.

Key findings

  • Successful file recovery with no bit errors required an average of 10 copies per sequence (SD 3), regardless of file size or pool complexity; the files encoded 0.1 KB (2,042 sequences), 1.7 KB (26,404 sequences), and 18 KB (271,447 sequences).
  • The demonstrated density was 17 EB g−1, nearly two orders of magnitude greater than prior work; with 15% logical redundancy the maximum estimated density was 38 EB g−1, and with 50% logical redundancy it was 124 EB g−1, ignoring other errors.
  • No distinguishable difference in sequence loss was found between water-diluted and 150 Nmer-diluted complex pools; for the large file, only three samples differed, with a mean difference of 0.98% missing sequences (SD 1.82%).
  • Increasing sequencing coverage from 20× to 200× reduced missing sequences by only 1.8% on average (SD 1.1%); samples successfully recovered at the lowest copy number had mean sequencing coverage of 35× (SD 11×).

Methods

  • Three files were randomly accessed from a DNA pool encoding nine files (1.6 M total oligonucleotides), then serially diluted either in water or in 1 ng μL−1 150 Nmers to raise complexity from 1.6 × 10^6 to over 10^10 unique sequences per microliter.
  • Copy numbers were measured by triplicate qPCR; for small and medium files, diluted-sample copy numbers were calculated using the large file’s dilution factors because of amplification-efficiency differences.
  • Sequencing data were randomly sub-sampled with replacement to 20× coverage 100 times per file per sample to estimate missing sequences; sequence loss was modeled with power regressions (R² 0.90–0.98).

Limitations

  • The paper states that appropriate logical and physical redundancy depends on the specific DNA data storage workflow, since synthesis, sequencing, and wet-protocol errors can affect sequence recovery.
  • The paper states that only one of the two dilution series was sequenced, and that small and medium file copy numbers for diluted samples were inferred from the large file’s dilution factors rather than measured directly by qPCR.
  • The paper states that maximum density estimates of 38 and 124 EB g−1 ignore all other errors such as insertions, deletions, and substitutions; it also notes that increasing sequencing coverage is useful only as a last resort because of cost.
  • No dedicated limitations section is present; the above are limitations stated in the text, not an external critique.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1038/s41467-020-14319-8

Cites

21 of 21 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Claims the paper makes

0 of 13 cited sources checked · 0 supported · 0 flagged

  • Findingown data

    Many sequences absent from the initial pool reappear in subsequent dilutions.

    “many of the strands “reappear” in subsequent dilutions as shown briefly in Fig. 3c”
    Quote in the paperSubstantial claim
  • Findingreasoning

    The small file lost fewer sequences than the medium and large files after copy number fell below ~10.

    “However, though the medium and large files behaved almost indistinguishably, the small file did not lose sequences at a similar rate after the copy number fell below ~10 (see Fig. 2a). Instead, it lost fewer sequences than the larger two files.”
    Quote in the paperNot a substantial claim
  • Findingown data

    Increasing sequencing coverage from 20× to 200× resulted in a mean of 1.8% fewer missing sequences.

    “This resulted in a mean of 1.8% fewer missing sequences for 200× coverage, with a standard deviation of 1.1% (Fig. 3b).”
    Quote in the paperSubstantial claim
  • Findingown data

    The same data can be successfully retrieved from a complex pool of over 10^10 unique sequences per microliter.

    “We successfully retrieve the same data in a complex pool of over 1010 unique sequences per microliter with no evidence that we have begun to approach complexity limits.”
    • Also stated in: § Reducing copy number and increasing pool complexity · plain
      “By diluting samples with 1 ng μL−1 150 Nmers, pool complexity increased from 1.6 × 106 unique sequences in the initial pool (Fig. 1b) to over 1010 unique sequences per microliter (see Supplementary Note 2).”
    Quote in the paperSubstantial claim
  • Findingown data

    No distinguishable difference in sequence loss was observed between water dilution and complex 150 Nmer dilution conditions.

    “We found no distinguishable difference when comparing sequence loss between water and complex 150 Nmer dilution conditions (Fig. 2a).”
    Quote in the paperSubstantial claim
  • May combine two claims. 14
  • Classified as probably not a substantial claim. 11

  • May combine two claims.

    Findingown data

    The samples successfully recovered at the lowest possible copy number had a mean sequencing coverage of 35× with a standard deviation of 11×, with a mean copy number of 10 and a standard deviation of 3.

    “We found that the samples successfully recovered at the lowest possible copy number had a mean sequencing coverage of 35× with a standard deviation 11×, with a mean copy number of 10 and a standard deviation of 3 when accounting for all files and all diluent conditions (Supplementary Note 7, Fig. 3a).”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingreasoning

    The storage density achieved is about 17 exabytes/gram, nearly two orders of magnitude greater than prior work.

    “This results in density of about 17 exabytes/gram, nearly two orders of magnitude greater than prior work has shown.”
    • Also stated in: § Data density · plain
      “Determining the need for a minimum of ~10 copies of each DNA sequence in the PCR reaction to successfully recover a file enables us to calculate a density of 17 EB g−1, nearly two orders of magnitude denser than prior work7 and closer to the maximum theoretical density predicted for DNA data storage1 (see Supplementary Note 5).”
    • Also stated in: § Discussion · boosted
      “By demonstrating files with a copy number of approximately 10 can be successfully recovered, we present the most practically dense system to date at 17 EB g−1, nearly two orders of magnitude greater than the densest prior work7.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [7] Erlich, Y. & Zielinski, D. DNA Fountain enables a robust and efficient storage architecture. Science 355, 950–954 (2017). source

      Not checked in the free run.

  • May combine two claims.

    Findingreasoning

    The encoding scheme and experimental protocol emulate ~150 GB of digital data per microliter, and 7 TB in the final 50 μL solution.

    “At ~15.6 bytes of data encoded per strand10, this encoding scheme and experimental protocol emulate ~150 GB of digital data per microliter, and 7 TB in the final 50 μL solution.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [10] Organick, L. et al. Random access in large-scale DNA data storage. Nat. Biotechnol. 36, 242–248 (2018). source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    The human and mouse genomes are both nearly 3 gigabases.

    “Though the commonly analyzed human and mouse genome are both nearly 3 gigabases13, approximately two orders of magnitude fewer bases than examined in this work’s most complex condition, genomic libraries are often exceedingly complex because of the genome fragmentation process.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [13] Geer, L. Y. et al. The NCBI BioSystems database. Nucleic Acids Res. 38, D492–D496 (2010). source

      Not checked in the free run.

  • May combine two claims.

    Interpretationreasoning

    The encoding scheme used in this work has 15% logical redundancy, which tolerates a maximum of 13% of sequences missing.

    “The logical redundancy used in this work was 15%, which tolerates a maximum of 13% of sequences missing (if there are 100 sequences and we apply 15% logical redundancy, there are now 100+15 sequences and we can tolerate 15/115 missing).”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationown data

    A relatively small increase in logical redundancy allows for a disproportionately lower physical redundancy with no data loss, resulting in higher overall data density.

    “A relatively small increase in logical redundancy allows for a disproportionately lower physical redundancy with no data loss. This results in higher overall data density.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Interpretationreasoning

    Both factors likely minimize interactions between the desired strands and the strands added for complexity.

    “Both factors likely minimize interactions between the desired strands and the strands added for complexity.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationown data

    The process examined here is robust and has high data density.

    “These findings substantiate the robustness and high data density of the process examined here.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationown data

    Sequence loss is stochastic and not due to sequence design.

    “find no systematic sequence loss, thus showing sequence loss is stochastic and not due to sequence design.”
    • Also stated in: § Sequence behavior · hedged
      “Thus, sequences may be changing proportions or are observed to be missing due to stochastic variation in sub-sampling.”
    • Also stated in: § Sequence behavior · hedged
      “this indicates that these sequences disappeared and reappeared due to stochastic effects rather than systematic interactions that made certain sequences irretrievable.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    A different method of random access using bead extraction utilized 10^18 unique sequences.

    “In more recent work, a different method of random access using bead extraction of desired strands prior to PCR random access utilized 1018 unique sequences11.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [11] Tomek, K. J. et al. Driving the scalability of DNA-based information storage systems. ACS Synth. Biol. 8, 1241–1248 (2019). source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    The effect of homopolymers in the payload has recently been found to be minimal in practice.

    “While the effect of homopolymers in the payload has recently found to be minimal in practice19, it is encouraging to note that the care taken to randomize payloads yields no systematic sequence loss.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [19] Chen, Y.-J. et al. Quantifying molecular bias in DNA data storage. Preprint at https://www.biorxiv.org/content/10.1101/566554v1 (2019). source

      Not checked in the free run.

  • May combine two claims.

    Interpretationown data

    The maximum data density is 38 EB g−1 for the scheme with 15% logical redundancy and 124 EB g−1 for 50% logical redundancy, based on the rate at which sequences are lost and ignoring all other errors.

    “Based on the rate at which sequences are lost (Fig. 2b) and ignoring all other errors (such as insertions, deletions, substitutions), the maximum data density is 38 EB g−1 for the scheme with 15% logical redundancy and 124 EB g−1 for 50% logical redundancy.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    The encoding scheme presented in prior work encodes the payload between primer regions using a randomized seed to generate a random payload and exclude homopolymers.

    “It is important to note that the encoding scheme presented in prior work10 and used here encodes the payload between the primer regions by using a randomized seed to generate a random payload and exclude homopolymers.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [10] Organick, L. et al. Random access in large-scale DNA data storage. Nat. Biotechnol. 36, 242–248 (2018). source

      Not checked in the free run.

  • May combine two claims.

    Interpretationown data

    The storage method used in this work enables efficient, reliable data retrieval in complex settings.

    “we demonstrate the ability of the storage method used in this work to enable efficient, reliable data retrieval in complex settings.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Interpretationreasoning

    The methods of prior work differ substantially from the methods presented here and make it difficult to compare.

    “However, in addition to different random access techniques and strand architecture, those methods differ substantially from the methods presented here and make it difficult to compare to this work.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    The mutagenetic PCR method introduces an approximate 5% error rate to the final PCR product.

    “their methods employ mutagenetic PCR on template strands which introduces an approximate 5% error rate to the final PCR product12 while maintaining a conserved PCR primer region.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [12] Zaccolo, M. & Gherardi, E. The effect of high-frequency random mutagenesis on in vitro protein evolution: a study on tem-1 β-lactamase. J. Mol. Biol. 285, 775–783 (1999). source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Interpretationcitation

    The limit to random access complexity is not 10^10 sequences per random access reaction.

    “Nevertheless, both works are consistent in concluding that the limit to random access complexity is not 1010 sequences per random access reaction.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Interpretationreasoning

    The method of increasing sequencing coverage has the benefit of not requiring additional material from the original pool because there is extra material post library preparation to sequence many times over.

    “This method of increasing sequencing coverage has the benefit of not requiring additional material from the original pool because there is extra material post library preparation to sequence many times over.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationown data

    The finding that sequences are lost stochastically coupled with no recovery threshold difference between water and complex settings assures users that the limit of pool complexity has not been reached.

    “The finding that sequences are lost stochastically coupled with the finding that there is no recovery threshold difference between samples diluted with water and those in more complex settings diluted with 150 Nmers thus assures users of this encoding scheme that we have not yet reached the limit of how complex a pool can be before it inhibits the ability to recover desired information.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Interpretationreasoning

    Determining the appropriate logical and physical redundancy depends on the specific DNA data storage workflow, for errors incurred by DNA synthesis, sequencing, and wet protocols used can affect sequence recovery.

    “It is important to note that determining the appropriate logical and physical redundancy depends on the specific DNA data storage workflow, for errors incurred by DNA synthesis, sequencing, and wet protocols used can affect sequence recovery.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationreasoning

    Theoretically, a duplicate sequence would be encountered only after 4^150 sequences had been synthesized; thus, each strand is expected to be unique.

    “Theoretically, we would expect to encounter a duplicate sequence only after 4150 (2 × 1090) sequences had been synthesized; thus, we would expect each strand to be unique.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Interpretationreasoning

    The small file's lower sequence loss after copy number fell below ~10 is likely due to a combination of copy number being slightly higher than calculated, fewer sequences initially missing due to variation in synthesis, and the distribution of sequences being slightly more uniform.

    “This is likely due to a combination of copy number being slightly higher than calculated, fewer sequences initially missing due to variation in synthesis, and the distribution of sequences being slightly more uniform (see Supplementary Note 3 for detailed analysis).”
    Quote in the paperNot a substantial claim
22 of 22

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

  • The storage density of DNA data storage increased from 2.2 PB/g to 17,000 PB/g.

    “The storage density has increased from 2.2 PB/g (petabytes per gram DNA) to 17,000 PB/g (Erlich and Zielinski, 2017; Organick et al., 2020).”

    From Large-Scale de Novo Oligonucleotide Synthesis for Whole-Genome Synthesis and Data Storage: Challenges and Opportunities (2021) · cites this paper as Organick et al., 2020

  • DNA has a theoretical data density of 6 bits for every 1 nm of polymer, or ∼4.5 × 10^7 GB/g, which is orders of magnitude higher than the densities achievable using traditional devices.

    “On the basis of its physical dimensions, DNA has a theoretical data density of 6 bits for every 1 nm of polymer, or ∼4.5 × 107 GB/g,6 which is orders of magnitude higher than the densities achievable using traditional devices.7,8”

    From Emerging Approaches to DNA Data Storage: Challenges and Prospects (2022) · cites this paper as [6]

  • The maximum data that can be stored within a single macroscopic storage container has been estimated to lie between a few TB up to a few hundred TB.

    “This represents the maximum data that can be stored within a single macroscopic storage container, and has been estimated to lie between a few TB up to a few hundred TB.56,6,70”

    From Emerging Approaches to DNA Data Storage: Challenges and Prospects (2022) · cites this paper as [6]

  • Currently, the highest demonstrated data capacity for random access is on the order of terabytes of data.

    “Currently, the highest demonstrated data capacity for random access is on the order of terabytes of data.6,12,71”

    From Emerging Approaches to DNA Data Storage: Challenges and Prospects (2022) · cites this paper as [6]

  • It is likely unpractical to scale random access by PCR-based addressing indefinitely because of the difficulty of orthogonal primer design and the requirement for many amplification cycles given PCR bias.

    “While this does not appear to be a hard limit,6 it is likely unpractical to scale random access by PCR-based addressing indefinitely because of the aforementioned difficulty of orthogonal primer design and the requirement for many amplification cycles given the associated impact of PCR bias.20”

    From Emerging Approaches to DNA Data Storage: Challenges and Prospects (2022) · cites this paper as [6]

Cites

Flagged first, then by citations

Cited by

Sorted by citations

Related

Authors
Organick, L., Chen, Y.-J., Dumas Ang, S., Lopez, R., Liu, X., Strauss, K., & Ceze, L.
Affiliations
  • University of Washington (US)
  • Microsoft (United States) (US)
Corresponding author
  • Organick, Lee
  • Strauss, Karin
ORCID
Venue
Nature Communications
Publisher
Springer Science and Business Media LLC
Published
2020-01-30
Volume
11
Issue
1
Pages
616-616
ISSN
2041-1723 (ISSN-L)
DOI
10.1038/s41467-020-14319-8
PMC
PMC6992699
OpenAlex
W3003918453
Semantic Scholar
a5133ffd70b88f4999335667a3014d88e383353c
Cited by
135
References
21
Citations by year
2026: 25 · 2025: 20 · 2024: 24 · 2023: 13 · 2022: 16 · 2021: 26 · 2020: 11
FWCI
5.35
Type
journal article
Language
en
Update notices
correction (2020-02-21)
Open access
Licence
Locations
  • Nature Communications (publishedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • RePEc: Research Papers in Economics (submittedVersion)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion)
Funders
  • Microsoft
  • United States Department of Defense | Defense Advanced Research Projects Agency
Topics
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › DNA and Biological Computing
  • Physical Sciences › Computer Science › Computer Networks and Communications › Advanced Data Storage Technologies
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Advanced biosensing and bioanalysis techniques
Keywords
DNA data storage, synthetic DNA, data archiving, file recovery, data density, sequencing depth, nucleotide sequence
MeSH terms
Base Sequence, DNA, Genetic Engineering, Information Science, Information Storage and Retrieval, Gene Dosage, Databases, Nucleic Acid, High-Throughput Nucleotide Sequencing

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Claims read by deepseek-v4.1-flash under definition version 3 (extract-v3) on 2026-10-09.

Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-09.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review