CoChat
Sign Up
PreprintCited by 110DNA and Biological ComputingAlgorithms and Data CompressionAdvanced Data Storage Technologies

Quantifying Molecular Bias in DNA Data Storage

Open access
Chen, Yuan-Jyue, Takahashi, Christopher N., Organick, Lee, Stewart, Kendall, Ang, Siena Dumas, Weiss, Patrick, Peck, Bill, Seelig, Georg +2 more·2019·Nature Communications
DOI 10.1101/566554PMC PMC7324401

Preprint — published as 10.1038/s41467-020-16958-3 · Paper page

Open this paper in CoChat — free accountOpen PDF

Abstract

DNA has recently emerged as an attractive medium for future digital data storage because of its extremely high information density and potential longevity. Recent work has shown promising results in developing proof-of-principle prototype systems. However, very uneven (biased) sequencing coverage distributions have been reported, which indicates inefficiencies in the storage process and points to optimization opportunities. These deviations from the average coverage in oligonucleotide copy distribution result in sequence drop-out and make error-free data retrieval from DNA more challenging. The uneven copy distribution was believed to stem from the underlying molecular processes, but the interplay between these molecular processes and the copy number distribution has been poorly understood until now. In this paper, we use millions of unique sequences from a DNA-based digital data archival system to study the oligonucleotide copy unevenness problem and show that two important sources of bias are the synthesis process and the Polymerase Chain Reaction (PCR) process. By mapping the sequencing coverage of a large complex oligonucleotide pool back to its spatial distribution on the synthesis chip, we find that significant bias comes from array-based oligonucleotide synthesis. We also find that PCR stochasticity is another main driver of oligonucleotide copy variation. Based on these findings, we develop a statistical model for each molecular process as well as the overall process and compare the predicted bias with our experimental data. We further use our model to explore the trade-offs between synthesis bias, storage physical density and sequencing redundancy, providing insights for engineering efficient, robust DNA data storage systems.

AI snapshot

Written from the paper’s full text.

TL;DR

Using UMI labeling, a ready-to-sequence 1,536,168-sequence pool, and dilution-PCR experiments, this study identifies DNA synthesis and PCR stochasticity as the two main sources of biased oligonucleotide copy distributions in DNA data storage. It builds a computational model of synthesis, PCR, and sequencing that predicts bias and explores trade-offs among synthesis bias, physical density, logical redundancy, and sequencing redundancy.

Key findings

  • Synthesis is a prominent bias source: a UMI-labeled DNA file with over 400,000 sequences (sample size 457,772) showed biased UMI counts after synthesis, normalized to 7.7 mean coverage, while read counts were normalized to 83.0 mean coverage. Synthesis bias was related to spatial location on the synthesis chip; the first 1,536,168-sequence ready-to-sequence pool did not fit a normal distribution, but an improved synthesis process did.
  • PCR stochasticity, not GC content, dominated PCR bias: the coefficient of variation before and after 31 cycles of PCR was 0.41 vs. 0.45 when subsampled to 20× coverage, and the standard deviation of population fraction change was 0.24 before PCR vs. 0.37 after PCR (p < 0.005, bootstrapping n = 1000). GC-content slopes were <0.01 for both the non-homopolymer pool (1,536,168 sequences, GC 40–60%) and homopolymer pool (1,358,998 sequences, GC 25–75%).
  • In dilution-PCR experiments with 7,373 sequences, average copy numbers from 8 to 113, and 18 PCR cycles, lower copy number increased skew; the model predicted sequencing-coverage c.v. with R² = 0.71 and population-fraction-change standard deviation with R² = 0.84.
  • Model trade-offs: storing as few as 10 copies per oligo could achieve <2% sequence dropout and a physical density of 9.3 EB/g; at 30 copies the required logical redundancy was 3%, whereas at 10 copies it grew to 8%, nearly tripling physical density. Biased synthesis (high c.v.) was the dominant factor in sequence dropout and could not be proportionally compensated by additional sequencing reads.

Methods

  • UMI labeling was applied to an arbitrarily chosen DNA archival file with over 400,000 sequences; after ligation, PCR enrichment, sequencing, and BWA alignment, UMI filtering was used to decouple synthesis bias from PCR bias.
  • A ready-to-sequence Twist Bioscience pool with 1,536,168 unique sequences was sequenced directly and after PCR (one 6-cycle plus five 5-cycle PCR processes, total 31 cycles); a separate homopolymer pool with 1,358,998 sequences was used to test GC-content effects on population fraction change.
  • A serial dilution-PCR experiment used a 7,373-sequence pool diluted to average copy numbers of 8–113, amplified with 18 PCR cycles, and sequenced; the computational model used a normal distribution for synthesis copy numbers, a stochastic branching process for PCR with P = 0.95, and random sampling for sequencing and dilution.

Limitations

  • Stated: experiments used Twist Bioscience synthesis and Illumina sequencing; applying the model to other technologies likely requires additional quantification and modeling, such as Ion Torrent or Oxford Nanopore homopolymer limitations and different array-based synthesis biases.
  • Stated: PCR experiments amplified a single file without other non-targeted files in the pool; future work is needed for complex pools, where non-specific primer binding may increase copy-number variation.
  • Stated/evident: the logical-redundancy example ignores insertions, deletions, and substitutions; the model treats high-copy-number PCR as effectively deterministic and models sequencing/dilution as random sampling, with synthesis modeled as normal for unbiased processes.

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

Integrity signals

Registry

Crossref · 10.1101/566554

Cites

35 of 35 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Claims the paper makes

18 of 26 cited sources checked · 12 supported · 0 flagged · accuracy not yet measured

  • Findingown data

    Synthesis bias is highly related to spatial location of oligos on a synthesis chip.

    “We find that synthesis bias is highly related to spatial location of oligos on a synthesis chip.”
    • Also stated in: another passage · hedged
      “indicating that synthesis bias was related to the spatial location on the synthesis chip.”
    Quote in the paperSubstantial claim
  • Findingown data

    There is no practically important association between GC content and PCR bias in DNA storage.

    “Even though these two encoding strategies lead to different GC distributions, we find no practically important association between GC content and PCR bias.”
    Quote in the paperSubstantial claim
  • Findingown data

    The old synthesis process is far more skewed than the improved process.

    “indicating that the old synthesis process is far more skewed (there have since been process improvements, discussed in the next section).”
    Quote in the paperNot a substantial claim
  • Findingown data

    The computational model's predicted bias is strongly associated with experimental data.

    “We observe strong association between the bias predicted from this model and from our experimental data.”
    Quote in the paperSubstantial claim
  • Findingown data

    There is significant bias from DNA synthesis.

    “UMI labeling allow us to decouple synthesis bias from PCR bias, and we find significant bias from DNA synthesis.”
    Quote in the paperSubstantial claim
  • Classified as probably not a substantial claim. 13
  • May combine two claims. 18
  • The quote only partly states this claim. 2

  • Classified as probably not a substantial claim.

    Premisecitation

    GC content of individual sequences affects PCR amplification efficiency in biological DNA.

    “GC content of individual sequences were previously found to affect PCR amplification efficiency in biological DNA15–17.”
    Quote in the paperSubstantial claim2 of 3 sources checked
    • cites [15] Aird, D. et al. Analyzing and minimizing PCR amplification bias in Illumina sequencing libraries. Genome Biol. 12, (2011). source
    • cites [16] source
    • cites [17] source

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

    Premisecitation

    Previous studies observed PCR bias in genomic biological sample amplification.

    “Although previous studies observed PCR bias in genomic biological sample amplification15–17, it remained unclear whether such bias is significant in DNA data storage.”
    Quote in the paperNot a substantial claim2 of 3 sources checked
    • cites [15] Aird, D. et al. Analyzing and minimizing PCR amplification bias in Illumina sequencing libraries. Genome Biol. 12, (2011). source
    • cites [16] source
    • cites [17] source

  • May combine two claims.

    Premisecitation

    Sequencing coverage of oligonucleotides in DNA data storage systems is very uneven, requiring modern error correction codes capable of handling sequence dropout.

    “However, the sequencing coverage (number of read counts of a unique sequence) of an oligonucleotide (henceforth referred to simply as “oligo”) was found to be very uneven, requiring modern error correction codes capable of handling sequence dropout7–12.”
    Quote in the paperSubstantial claim5 of 6 sources checked
    • cites [7] Organick, L. et al. Random access in large-scale DNA data storage. Nat. Biotechnol. 36, (2018). source
    • cites [8] source
    • cites [9] source

      Methods > Address Design and Encoding (page unavailable) says GC content affects sequencing coverage, and the Abstract (page unavailable) discusses coding against sequencing noise and assembly errors. The paper does not report very uneven read counts across oligonucleotides or say that sequence dropout requires dropout-tolerant error-correction codes. (LLM, conf=0.78)

      Experimental — accuracy not yet measured
    • cites [10] 2016, A DNA-Based Archival Storage System, ACM SIGOPS Oper. Syst. Rev source
    • cites [11] 2017, Portable and Error-Free DNA-Based Data Storage, Sci. Rep source
    • cites [12] source

      In System Implementation (Table 2; page unavailable), read counts vary substantially across the DNA blocks, supporting uneven coverage in this experiment. The Encoding Step and Post-processing Step describe codes for sequencing deletion and homopolymer errors, not codes capable of handling complete sequence dropout. The paper therefore does not support the claimed need for dropout correction. (LLM, conf=0.85)

      Experimental — accuracy not yet measured

  • May combine two claims.

    Premisecitation

    PCR is an exponential process, so small random variations early on in amplification can have a large impact on distribution.

    “PCR is an exponential process, so small random variations early on in amplification can have a large impact on distribution18–24.”
    Quote in the paperSubstantial claim3 of 7 sources checked
    • cites [18] Hassibi, A. , Kakavand, H. & Lee, T. H. A stochastic model and simulation algorithm for polymerase chain reaction (PCR) systems. Proc. GENSIPS (2004). source
    • cites [19] 2005, Confidence intervals for nonhomogeneous branching processes and polymerase chain reactions, Ann. Probab source
    • cites [20] source
    • cites [21] source

      The Model of the PCR (p. 3) describes random branching during PCR’s exponential phase, and the introduction (p. 2) reports significant finite-population effects and persistent variance. These passages support sensitivity to randomness, but they do not specifically establish that small variations early in amplification have a large impact on the final distribution. (LLM, conf=0.78)

      Experimental — accuracy not yet measured
    • cites [22] Kebschull, J. M. & Zador, A. M. Sources of PCR-induced distortions in high-throughput sequencing data sets. Nucleic Acids Res. 43, (2015). source
    • cites [23] source
    • cites [24] source

      The Materials and Methods sections on plasmids and chromatin immunoprecipitation describe using PCR to generate fragments and analyze samples. Neither those passages nor searches for exponential amplification and early-cycle stochastic variation address whether small random variations can substantially change the resulting distribution. Page numbers are unavailable in our copy. (LLM, conf=0.76)

      Experimental — accuracy not yet measured

  • May combine two claims.

    Study design and conductown data

    High-throughput sequencing and sample dilution are modeled using random sampling because GC content in the files did not show practically significant bias in the PCR GC bias test.

    “Illumina sequencing was previously observed to have bias on GC-extreme sequences15,27,28, but GC content in our files did not show practically significant bias in the PCR GC bias test. Therefore, high-throughput sequencing and sample dilution are modeled using random sampling.”
    Quote in the paperSubstantial claim3 of 3 sources checked
    • cites [15] Aird, D. et al. Analyzing and minimizing PCR amplification bias in Illumina sequencing libraries. Genome Biol. 12, (2011). source
    • cites [27] source
    • cites [28] source

  • May combine two claims.

    Interpretationown data

    The estimated physical density is over 10-fold higher than prior work by Erlich and Zielinski and is aligned with what has been observed in practice.

    “This estimated physical density is over 10-fold higher than prior work by Erlich and Zielinski10 and is aligned with what we have recently observed in practice29.”
    Quote in the paperSubstantial claim2 of 2 sources checked
    • cites [10] 2016, A DNA-Based Archival Storage System, ACM SIGOPS Oper. Syst. Rev source
    • cites [29] source

      The Results section on U1 levels reports cancer-cell migration and invasion, while the Methods section on LECDS calculations concerns RNA-seq analysis; neither addresses physical density. Searches for the tenfold comparison, Erlich and Zielinski, and physical density in practice found no relevant passage in our copy. (LLM, conf=0.78)

      Experimental — accuracy not yet measured

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

    Premisecitation

    Prior work found that PCR copy data from a deeply diluted oligo pool resulted in dramatic bias, which is less suitable for data recovery.

    “Indeed, prior work also found that PCR copy data from a deeply diluted oligo pool resulted in dramatic bias, which is less suitable for data recovery10.”
    Quote in the paperNot a substantial claim1 of 1 sources checked
    • cites [10] 2016, A DNA-Based Archival Storage System, ACM SIGOPS Oper. Syst. Rev source

      The Text reports that nine serial PCR amplifications produced a deep copy with substantially more skewed oligo representation than the master pool. But the decoder fully recovered the file from that copy, so the claim overstates its effect on recovery. The paper separately reports high dropout rates for libraries made from very small DNA inputs; those dilution results should not be conflated with the deep-copy result. (LLM, conf=0.86)

      Experimental — accuracy not yet measured

  • May combine two claims.

    Findingown data

    PCR stochasticity widens oligo copy distributions of the test DNA archival file and seems to be a dominant factor in PCR bias.

    “Instead, we find that PCR stochasticity widens oligo copy distributions of our test DNA archival file and, based on our observations, seems to be a dominant factor in PCR bias.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    The amplification ratio has high variation when oligos have very low copy numbers, indicating that the amplification ratio is affected by stochastic effects at these low copy numbers.

    “On the other hand, the amplification ratio was observed to have high variation when oligos had very low copy numbers, indicating that the amplification ratio was affected by stochastic effects at these low copy numbers.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    The distribution after synthesis is very similar in shape to the distribution after PCR, indicating that PCR does not significantly increase bias overall.

    “This distribution is also very similar in shape to the distribution after PCR, indicating that PCR does not significantly increase bias overall.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    The oligo distribution before synthesis process improvement did not follow a normal distribution, but the oligo distribution using the improved synthesis process is well fitted to a normal distribution.

    “Interestingly, the oligo distribution before the synthesis process improvement did not follow a normal distribution, but the oligo distribution using the improved synthesis process is now well fitted to a normal distribution (Fig. 3b, d).”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    At 30 copies, the required logical redundancy for data recovery is 3%, whereas at 10 copies the logical redundancy grows to only 8%, nearly tripling physical density.

    “For example, at 30 copies, the required logical redundancy for data recovery is 3% whereas at 10 copies the logical redundancy grows to only 8%, nearly tripling physical density.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    In the ready-to-sequence pool with GC content between 40 and 60%, the association between population fraction changes and GC content was statistically significant (P < 0.05) but very small and practically unimportant (slope < 0.01).

    “Although the association between population fraction changes and GC content of this pool (between 40 and 60%) was found to be statistically significant (P value < 0.05), the association between the two was very small and practically unimportant (the slope of the linear fit was <0.01).”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    A biased synthesis pool (high c.v.) is the dominant factor in sequence dropout and cannot be proportionally compensated by additional sequencing reads.

    “It shows that a biased synthesis pool (i.e., high c.v.) is the dominant factor in sequence dropout and cannot be proportionally compensated by additional sequencing reads.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    In 9 homopolymer DNA archival files with 1,358,998 unique sequences and GC content from 25 to 75%, the association between GC content and population fraction changes was very small and not practically important (slope < 0.01).

    “These homopolymer files had a wider range of GC content from 25 to 75%, but the association between GC content and the population fraction changes was still very small and not practically important (the slope of the linear fit was <0.01).”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    The standard deviations of population fraction changes were 0.24 before PCR and 0.37 after PCR, and these two numbers were statistically different (p < 0.005, bootstrapping n = 1000).

    “The standard deviations of population fraction changes were 0.24 and 0.37 before PCR and after PCR, respectively, and these two numbers were statistically different (p < 0.005, computed by bootstrapping n = 1000).”
    Quote in the paperSubstantial claim

  • The quote only partly states this claim.

    Study design and conductreasoning

    A ready-to-sequence pool with 1,536,168 unique sequences can be sequenced directly without intermediate PCR or DNA ligation steps required for sequencing library preparation.

    “This ready-to-sequence pool can be sequenced using an Illumina sequencer directly, with no need for intermediate PCR or DNA ligation steps required for sequencing library preparation.”
    Quote in the paperSubstantial claim

  • May combine two claims. The quote only partly states this claim.

    Premisereasoning

    Twist Bioscience increased the dosage tolerance window and introduced engineering changes, resulting in dramatically decreased error rates and more robust molecules in subsequent processes.

    “The combination of these changes has resulted in dramatically decreased error rates and more robust molecules in subsequent processes.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Premisenone given

    Initial data used in this paper were produced with a process that allowed process excursions outside of the dosage tolerance window.

    “Initial data used in this paper were produced with a process that allowed process excursions outside of the dosage tolerance window.”
    Quote in the paperSubstantial claim

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

    Interpretationnone given

    Even an unbiased PCR process will have nonzero population fraction change for finite sample sizes, and this should asymptotically decrease with the total number of reads.

    “It is worth noting that even an unbiased process will have for finite sample sizes. Furthermore, should asymptotically decrease with the total number of reads.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S4 (based).

    Study design and conductnone given

    UMI labeling ensures with high probability that each molecule will include a tag different from any other.

    “Synthetic DNA pools include multiple copies of each sequence, and UMI labeling ensures with high probability that each molecule will include a tag different from any other.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationown data

    Oligos from unbiased synthesis and sequencing processes can be well modeled as a normal distribution and random sampling, respectively.

    “In this study, we found that oligos from unbiased synthesis and sequencing processes can be well modeled as a normal distribution and random sampling, respectively.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Premisereasoning

    PCR is not perfect; replication of an individual molecule has a probability of less than one, so even small random divergence in early phases of amplification can create significant bias known as PCR stochastic bias.

    “Because PCR is not perfect (i.e., replication of an individual molecule has a probability of less than one), even small random divergence in early phases of amplification can create significant bias, which is known as PCR stochastic bias.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Study design and conductnone given

    The experiment was designed to avoid complexity from other files for proper quantification of the impact of PCR stochastic bias.

    “This experiment was designed to avoid complexity from other files for proper quantification of the impact of PCR stochastic bias.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Study design and conductnone given

    The system model was experimentally tested by PCR-amplifying a single file without any other non-targeted files in a pool.

    “Our system model was experimentally tested by PCR-amplifying a single file without any other non-targeted files in a pool.”
    Quote in the paperSubstantial claim

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

    Study design and conductreasoning

    The example ignores all other errors such as insertions, deletions, and substitutions, which should be taken into account when determining proper logical redundancy.

    “It is worth pointing out the example here ignores all other errors such as insertions, deletions, and substitutions. These errors depend on synthesis and sequencing technologies, and they should be taken into account when determining the proper logical redundancy.”
    Quote in the paperNot a substantial claim

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

    Interpretationnone given

    The proposed system model can in principle be applied to other synthesis and sequencing technologies.

    “While the experiments were tested using Twist Bioscience and Illumina sequencing, the proposed system model can in principle be applied to other synthesis and sequencing technologies.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationcitation

    The negligible bias impact from GC content in the experimental data was likely because the oligos were relatively short (150-nt) and the use of KAPA HIFI polymerase reduced the impact of GC bias.

    “The negligible bias impact from GC content in our experimental data was likely because these oligos were relatively short (150-nt), and the use of KAPA HIFI polymerase also reduced the impact of GC bias26.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [26] Benjamini, Y. & Speed, T. P. Summarizing and correcting the GC content bias in high-throughput sequencing. 40, 1–14 (2012). source

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

    Interpretationnone given

    Different array-based synthesis technologies could have their own unique dependent bias caused by processes specific to them, such as uneven fluidic operation, surface treatment, and other factors.

    “Different array-based synthesis technologies could also have their own unique dependent bias caused by processes specific to them, such as uneven fluidic operation, surface treatment, and other factors.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing).

    Interpretationreasoning

    When applying the model to other technologies, additional quantification and modeling is likely needed.

    “It is worth noting that when applying our model to other technologies, additional quantification and modeling is likely needed.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S4 (based).

    Interpretationreasoning

    PCR carried out at high copy number of templates should obey the law of mass action and therefore be effectively deterministic.

    “PCR carried out at high copy number of templates should obey the law of mass action and therefore be effectively deterministic.”
    Quote in the paperNot a substantial claim
31 of 31

  • The paper relies on this without support.

    Current methods typically require either trial-and-error of experimental protocols or brute-force use of hundreds to thousands of sequencing reads per sequence to capture underrepresented sequences.

    “Current methods typically require either trial-and-error of experimental protocols or brute-force use of hundreds to thousands of sequencing reads per sequence to capture underrepresented sequences.”

  • The paper relies on this without support.

    The inefficiency in DNA data storage stems from lack of understanding about bias in oligo copy distribution and how it changes as oligos are manipulated in DNA data storage systems.

    “This inefficiency stems from lack of understanding about bias in oligo copy distribution, as well as how it changes as the oligos are manipulated in DNA data storage systems.”

  • The paper relies on this without support.

    Ion Torrent and Oxford Nanopore sequencing show limited ability to accurately sequence long homopolymers, which is less significant in Illumina sequencing.

    “For example, Ion Torrent and Oxford Nanopore sequencing show limited ability to accurately sequence long homopolymers, which is less significant in Illumina sequencing.”
3 of 3

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 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.”

    From Probing the physical limits of reliable DNA data retrieval. (2020) · cites this paper as [19]

  • PCR-based random access irreversibly removes oligos from the pool, which necessitates potentially lossy reamplification of the entire pool after repeated data retrieval.

    “Second, PCR-based random access irreversibly removes oligos from the pool, which necessitates potentially lossy reamplification of the entire pool after repeated data retrieval.75,76”

    From Emerging Approaches to DNA Data Storage: Challenges and Prospects (2022) · cites this paper as [76]

  • Each PCR amplification may introduce stochastic variation in copy number of the file sequences, leading to up to 2% data loss per amplification if using ten-fold physical redundancy.

    “In this case, each PCR amplification may introduce stochastic variation in copy number of the file sequences, leading to up to 2% data loss per amplification19 if using ten-fold physical redundancy, as recently suggested18.”

    From Random access DNA memory using Boolean search in an archival file storage system (2021) · cites this paper as [19]

  • The file system overcomes several challenges associated with pre-existing PCR-based file systems, including obviating the need for numerous heating and cooling cycles and enzymatic synthesis, and eliminating non-specific crosstalk between file sequences and barcodes, while enabling arbitrary Boolean logical search queries.

    “Our file system overcomes several challenges associated with pre-existing PCR-based file systems, including obviating the need for numerous heating and cooling cycles and enzymatic synthesis, and eliminating non-specific crosstalk between file sequences and barcodes17,19,21, while enabling arbitrary Boolean logical search queries.”

    From Random access DNA memory using Boolean search in an archival file storage system (2021) · cites this paper as [19]

  • Random access of data using PCR can cause data loss owing to biased amplification.

    “Random access of data using PCR can also cause data loss owing to biased amplification (28, 29).”

    From Electrochemical DNA synthesis and sequencing on a single electrode with scalability for integrated data storage (2021) · cites this paper as [28]

Cites

Flagged first, then by citations

Research works printed but not linked: 1

  1. Hassibi, A. , Kakavand, H. & Lee, T. H. A stochastic model and simulation algorithm for polymerase chain reaction (PCR) systems. Proc. GENSIPS (2004).

Cited by

Sorted by citations

Related

Authors
Chen, Y.-J., Takahashi, C. N., Organick, L., Stewart, K., Ang, S. D., Weiss, P., Peck, B., Seelig, G., Ceze, L., & Strauss, K.
Affiliations
  • Microsoft (United States) (US)
  • University of Washington (US)
  • Twist Bioscience (United States) (US)
Corresponding author
  • Chen, Yuan-Jyue
  • Strauss, Karin
ORCID
Venue
Nature Communications
Publisher
openRxiv
Published
2019-03-04
Volume
11
Issue
1
Pages
3264-3264
ISSN
2041-1723 (ISSN-L)
DOI
10.1101/566554
PMC
PMC7324401
OpenAlex
W3037026856
Cited by
109
References
35
Citations by year
2026: 14 · 2025: 22 · 2024: 15 · 2023: 19 · 2022: 13 · 2021: 15 · 2020: 8 · 2019: 3
FWCI
3.87
Type
journal article
Language
en
Open access
Licence
cc-by
Locations
  • Nature Communications (publishedVersion, cc-by)
  • bioRxiv (Cold Spring Harbor Laboratory) (acceptedVersion, cc-by-nd)
  • PubMed (publishedVersion)
  • PubMed Central (submittedVersion, cc-by)
  • RePEc: Research Papers in Economics (submittedVersion)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
Funders
  • U.S. Department of Defense
  • Microsoft
  • Defense Advanced Research Projects Agency
Topics
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › DNA and Biological Computing
  • Physical Sciences › Computer Science › Artificial Intelligence › Algorithms and Data Compression
  • Physical Sciences › Computer Science › Computer Networks and Communications › Advanced Data Storage Technologies
Keywords
DNA data storage, PCR bias, storage density, data archiving
MeSH terms
Models, Theoretical, Bias, Information Storage and Retrieval, Sequence Analysis, DNA

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.

Cited sources are checked by gpt-6-sol.

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