CoChat
Sign Up
Cited by 45DNA and Biological ComputingQuantum Information and CryptographyComputability, Logic, AI Algorithms

DNA synthesis for true random number generation

Open access
Meiser, Linda C., Koch, Julian, Antkowiak, Philipp L., Stark, Wendelin J., Heckel, Reinhard, Grass, Robert N.·2020·Nature Communications
DOI 10.1038/s41467-020-19757-yPMC PMC7675991
Open this paper in CoChat — free accountOpen PDF

Abstract

The volume of securely encrypted data transmission required by today’s network complexity of people, transactions and interactions increases continuously. To guarantee security of encryption and decryption schemes for exchanging sensitive information, large volumes of true random numbers are required. Here we present a method to exploit the stochastic nature of chemistry by synthesizing DNA strands composed of random nucleotides. We compare three commercial random DNA syntheses giving a measure for robustness and synthesis distribution of nucleotides and show that using DNA for random number generation, we can obtain 7 million GB of randomness from one synthesis run, which can be read out using state-of-the-art sequencing technologies at rates of ca. 300 kB/s. Using the von Neumann algorithm for data compression, we remove bias introduced from human or technological sources and assess randomness using NIST’s statistical test suite.

AI snapshot

Written from the paper’s full text.

TL;DR

The paper demonstrates true random number generation by synthesizing DNA strands with random nucleotides and reading them out by next-generation sequencing. After correcting synthesis biases with the von Neumann algorithm, the resulting bit streams pass NIST randomness tests and can yield large volumes of randomness, though sequencing remains a cost and speed bottleneck.

Key findings

  • Three commercial random-DNA syntheses showed two main biases: nucleotide nonequivalence (higher G and T than A and C) and position nonequivalence (G decreases and T increases from 5′ to 3′), plus nucleotide binding prevalence effects.
  • Von Neumann de-biasing removed the bias; de-biasing efficiency was 23.7% for Microsynth synthesis 1 and between 23% and 24% across syntheses.
  • After de-biasing, processed bit streams passed every NIST statistical test listed, with pass rates >54/56 and a required minimum of 52/56; P-values were ≥0.001.
  • From 204 µg of DNA, corresponding to about 4 × 10^15 strands, the authors report 7 PB of randomness after von Neumann correction, a synthesis rate of 225 GB/s at 0.000014 USD/GB, and sequencing readout at ca. 300 kB/s at 600 USD/GB using a full NovaSeq 6000 run.

Methods

  • DNA was designed with a 64-nucleotide random region flanked by primer regions, total length 105 nucleotides; three orders were placed with Microsynth (syntheses 1 and 2) and Eurofins Genomics (synthesis 3).
  • Samples were sequenced on Illumina iSeq 100 with paired-end 2 × 150 bp reads; sequences were filtered for correct adapters and length, then shortened to 60 nucleotides for analysis.
  • Nucleotides were mapped to bits as A → 0, C → 0, T → 1, G → 1; raw bit streams were de-biased using the von Neumann algorithm and evaluated with NIST’s statistical test suite.

Limitations

  • The paper states that limitations in cost and speed of random number generation by DNA synthesis, especially reading the DNA, will not vanish within the next months.
  • The von Neumann algorithm has a large data loss, discarding at least 75% of input bits, and low computational efficiency because output is about four times slower than input.
  • The paper states that synthesis and sequencing errors cannot be completely decoupled, making error bias difficult to distinguish; it also notes that coverage bias was checked and could not explain the observed nucleotide biases.
  • Evident from the text: the main-text NIST evaluation focuses on Microsynth synthesis 1, while results for other syntheses and mapping schemes are placed in Supplementary Information; only three commercial syntheses from two suppliers were analyzed.

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

Integrity signals

Registry

Crossref · 10.1038/s41467-020-19757-y

Cites

48 of 48 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Claims the paper makes

0 of 34 cited sources checked · 0 supported · 0 flagged

  • Findingown data

    Across all manufacturers, the curves from Microsynth synthesis 1 are much smoother than those from Microsynth 2 and Eurofins Genomics.

    “Across all manufacturers, the curves in Fig. 3 obtained from Microsynth synthesis 1 are much smoother than those obtained from Microsynth 2 and Eurofins Genomics syntheses.”
    Quote in the paperNot a substantial claim
  • Findingown data

    In all three commercial DNA syntheses, the percentage of G and T nucleotides is higher than the percentage of A and C in the random region.

    “In all syntheses, the percentage of G (guanine) and T (thymine) nucleotides is higher than the percentage of A (adenine), and C (cytosine) nucleotides. We call this trend nucleotide nonequivalence.”
    Quote in the paperSubstantial claim
  • Findingown data

    In all three syntheses, the percentage of A and C is relatively constant over the 60 nucleotides of the random region.

    “Whereas the percentage of A and C is relatively constant over the string of 60 nucleotides”
    Quote in the paperNot a substantial claim
  • Findingown data

    The position nonequivalence trend is stronger in both Microsynth syntheses than in the material from Eurofins.

    “This trend is stronger in both synthesis runs from Microsynth than in the material received from Eurofins.”
    Quote in the paperNot a substantial claim
  • Findingown data

    Computational efficiency of Von Neumann de-biasing is low, with average output rate four times slower than average input rate.

    “and computational efficiency is low (as the average output rate of data is four times slower than the average input rate of data)”
    Quote in the paperNot a substantial claim
  • May combine two claims. 24
  • Classified as probably not a substantial claim. 9

  • May combine two claims.

    Findingown data

    A single sequence is not present in the pool more than five times, with a mean presence of each sequence of 1.03 times.

    “We have analyzed this by counting the number of occurrences of each sequence and found that a single sequence is not present in the pool more than five times with a mean presence of each sequence of 1.03 times.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    In all three syntheses, the percentage of G decreases from 5′ to 3′ and the percentage of T increases from 5′ to 3′.

    “the percentage of G decreases from 5′ to 3′ and the percentage of T increases from 5′ to 3′.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    The nucleotide pair distribution cannot only be explained by the position-dependent bias; for example, CA should be less frequent than GG at the beginning of the sequence, yet they are equally likely.

    “Note that the nucleotide pair distribution cannot only be explained by the position-dependent bias as seen in Fig. 3a: For example, when just going by bias, CA should be less frequent than GG at the beginning of the sequence, yet they are equally likely.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Evaluation of raw bit streams using NIST showed that not all tests passed, implying the bit streams do not have the same statistical properties as a fully random sequence and still contain redundancy and bias.

    “Evaluation of the (raw) bit streams using NIST statistical test suite showed that not all tests passed, implying that the bit streams we obtain do not have the same statistical properties as a fully random sequence, i.e., they still contain some redundancy and bias”
    Quote in the paperSubstantial claim

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

    Findingnone given

    Using the von Neumann algorithm for data compression removes bias introduced from human or technological sources.

    “Using the von Neumann algorithm for data compression, we remove bias introduced from human or technological sources and assess randomness using NIST’s statistical test suite.”
    Quote in the paperNot a substantial claim

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

    Findingown data

    Microsynth synthesis 1 shows the most homogeneous mixing and coupling behavior during synthesis steps.

    “Although Microsynth synthesis 1 shows the strongest bias resulting from transversion, the smooth curves show the most homogeneous mixing and coupling behavior during synthesis steps.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingreasoning

    The dry DNA sample contains a theoretical entropy of 28 PB if there is no bias in the data, and 7 PB of randomness when removing bias using Von Neumann with a loss of 75% of bits.

    “The sample of dry DNA contains a theoretical entropy of 28 PB (if there is no bias in the data), and 7 PB of randomness when removing bias using Von Neumann with a loss of 75% of bits.”
    • Also stated in: § Calculations of cost and rate for randomness generation · plain
      “As a result, the 204 µg sample of dry DNA contain a theoretical entropy of 1 bits/nucleotide × 60 nucleotides/strand × 4 × 1015 strands = 28 PB (if there is no bias in the data), and 7 PB of randomness when calculating the ROV with 25% de-biasing efficiency.”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: There is no bias in the data for the theoretical entropy calculation of 28 PB.

  • May combine two claims.

    Findingown data

    DNA syntheses have two major biases, partly from synthesis discrepancies such as mixing and coupling efficiencies, and also from transversion originating from oxidation steps during DNA synthesis.

    “In this work, we have found that DNA syntheses have two major biases in part from synthesis discrepancies, such as mixing and coupling efficiencies, but also from transversion, originating from oxidation steps during DNA synthesis.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    The processed bit streams passed every NIST statistical test listed, with a pass rate for every test of >54/56, surpassing the statistically required minimum of 52/56.

    “As can be seen in Table 1, the processed bit streams passed every test listed with a pass rate for every test of >54/56, surpassing the statistically required minimum (52/56)12.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [12] Rukhin, A., Soto, J., Nechvatal, J. & Smid, M. A Statistical Test Suite for Random and Pseudorandom Number Generators for Cryptographic Applications. https://www.nist.gov/publications/statistical-test-suite-random-and-pseudorandom-number-generators-cryptographic (NIST, 2010).

      Not checked in the free run.

  • May combine two claims.

    Findingown data

    From Microsynth synthesis 1, 204 μg of DNA was obtained, translating to about 4 × 1015 strands of DNA.

    “From the synthesis by Microsynth (synthesis 1), we have obtained 204 µg of DNA, which translates to about 4 × 1015 strands of DNA.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    For a 2-bits-to-1-nucleotide mapping, the raw bit streams were not random and one Von Neumann compression did not make them random, but a second compression did make them random.

    “The raw bit streams were not random, and one Von Neumann compression did not make the bit streams random. However, after a second compression, bit streams did become random.”
    Quote in the paperSubstantial claim

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

    Premisecitation

    The Von Neumann generator has a large loss of data, as at least 75% of the input is discarded.

    “However, the big drawback of the Von Neumann generator is the large loss of data, as at least 75% of the input is being discarded. This implies that the input needs to be sufficiently large9,40.”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [9] Kwok, S.-H. et al. Information Security Theory and Practice. A Comparison of Post-Processing Techniques for Biased Random Number Generators (Springer, 2011). source

      Not checked in the free run.

    • cites [40] Dichtl, M. Bad and Good Ways of Post-processing Biased Physical Random Numbers (Springer, 2007).

      Not checked in the free run.

  • Classified as probably not a substantial claim.

    Premisecitation

    If data is handled accordingly, sequencing errors occur at random positions.

    “However, studies have suggested that if data is handled accordingly, sequencing errors occur at random positions35.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [35] Pfeiffer, F. et al. Systematic evaluation of error rates and causes in short samples in next-generation sequencing. Sci. Rep. 8, 1–14 (2018). source

      Not checked in the free run.

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

    Interpretationreasoning

    Adding more T than G building blocks to the reaction as synthesis proceeds could leverage the bias from transversion.

    “For example, adding more T than G building blocks to the reaction as synthesis proceeds (and thus altering the predetermined ratio of nucleotides A, G, T, and C) could leverage the bias from transversion.”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: Oxidation can lead to G–T transversion, whereby G is chemically altered such that during DNA replication it can be exchanged for T.

  • May combine two claims.

    Premisecitation

    Although expectation of products can be statistically predicted, being able to identify individual molecules after synthesis is rarely possible.

    “Although the expectation of products can be statistically predicted, being able to identify individual molecules after synthesis is rarely possible15.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [15] Lee, E. C., Parrilla-Gutiérrez, J. M., Henson, A., Brechin, E. K. & Cronin, L. A crystallization robot for generating true random numbers based on stochastic chemical processes. Matter 2, 1–9 (2020). source

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    Errors may be related to sequencing context, with highest rate for nucleotide T, which in our case would be erroneous once in every 1250 nucleotides, and error rates for C, G, and A being much lower.

    “Other studies have shown that errors may be related to the sequencing context, where the highest rate of errors was found for nucleotide T, which in our case would result to be erroneous once in every 1250 nucleotides, and the error rates for C, G, and A being much lower39.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [39] Schirmer, M., D’Amore, R., Ijaz, U. Z., Hall, N. & Quince, C. Illumina error profiles: resolving fine-scale variation in metagenomic sequencing data. BMC Bioinforma. 17, 1–15 (2016). source

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    Not being able to identify individual molecules results in loss of randomness when analyzing stochastic chemical processes, which is why chemical reactions cannot typically be used as RNGs.

    “Although this is a promising approach, not being able to identify individual molecules results in the loss of randomness when analyzing stochastic chemical processes, which is why chemical reactions cannot typically be used as RNGs15.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [15] Lee, E. C., Parrilla-Gutiérrez, J. M., Henson, A., Brechin, E. K. & Cronin, L. A crystallization robot for generating true random numbers based on stochastic chemical processes. Matter 2, 1–9 (2020). source

      Not checked in the free run.

  • May combine two claims.

    Interpretationreasoning

    By synthesizing 204 μg of DNA, random numbers can be synthesized at a rate higher than 225 GB/s, offering volumes of up to 7 million GB of randomness for a cost of 0.000014 USD/GB for synthesis and a fully scalable read-out on demand using Illumina sequencing technology.

    “By synthesizing 204 µg of DNA, we have shown the possibility of synthesizing random numbers at a rate higher than 225 GB/s, offering volumes of up to 7 million GB of randomness for a cost of 0.000014 USD/GB (synthesis) and a fully scalable read-out on demand using Illumina sequencing technology.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Premisecitation

    Previous work theoretically proposed a scheme for a possible automated workflow for DNA random number generation, but the physical realization of the theory and experimental limitations were not investigated.

    “Previous work has presented the idea of simulation of the DNA random number generation circuitry by theoretically proposing a scheme for a possible automated workflow for DNA random number generation. However, the physical realization of the theory and the experimental limitations were not investigated21,22.”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites [21] Gearheart, C. M., Arazi, B. & Rouchka, E. C. DNA-based random number generation in security circuitry. Biosystems 100, 208–214 (2010). source

      Not checked in the free run.

    • cites [22] Bogard, C. M., Rouchka, E. C. & Arazi, B. Random number generation for DNA-based security circuitry. BMC Bioinforma. 9, P13 (2008). source

      Not checked in the free run.

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

    Interpretationreasoning

    If all nucleotides were unbiased, bundling three nucleotides together would give an expected de-biasing efficiency of 28.72%, slightly improving over 1-bit-to-1-nucleotide mapping.

    “If all nucleotides were unbiased, the expected efficiency of de-biasing is 28.72%, thus showing a slight improvement over 1-bit-to-1-nucleotide mapping.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim. May combine two claims.

    Study design and conductcitation

    The NIST tests were applied with the standard parameters given by the NIST statistical test suite, with specified parameters differing from set values.

    “The tests were applied with the standard parameters as given by the NIST statistical test suite12, with the following parameters differing from set values:”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [12] Rukhin, A., Soto, J., Nechvatal, J. & Smid, M. A Statistical Test Suite for Random and Pseudorandom Number Generators for Cryptographic Applications. https://www.nist.gov/publications/statistical-test-suite-random-and-pseudorandom-number-generators-cryptographic (NIST, 2010).

      Not checked in the free run.

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

    Study design and conductcitation

    NIST evaluation tests were chosen such that for each test, 56 bit streams containing 1096 bits were tested, except for the rank test, where 56 bit streams containing 100,000 bits each were tested; this variant was chosen and explained by Rojas et al.

    “NIST evaluation tests were chosen such that for each test, 56 bit streams containing 1096 bits were tested (with the exception of the rank test, where 56 bit streams containing 100,000 bits each were tested). This variant of statistical analyses was chosen and explained by Rojas et al.14.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [14] Gaviria Rojas, W. A. et al. Solution-processed carbon nanotube true random number generator. Nano Lett. 17, 4976–4981 (2017). source

      Not checked in the free run.

  • Classified as probably not a substantial claim. May combine two claims.

    Study design and conductreasoning

    To allow only sequences that were synthesized and sequenced correctly, sequences containing 16 nucleotides of the adapter were searched for and used for further data treatment, and all other sequences were discarded.

    “To only allow for sequences that have been synthesized and sequenced correctly, sequences containing 16 nucleotides of the adapter (AGA TCG GAA GAG CAC A) were searched for and used for further data treatment. All other sequences were discarded.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationreasoning

    DNA synthesis is not the bottleneck in DNA random number generation; a standard synthesis can generate true randomness at a rate of 225 GB/s at a cost of 0.000014 USD/GB.

    “Consequently, in contrast to DNA data storage38, DNA synthesis is not the bottleneck in DNA random number generation as such a standard synthesis can generate true randomness at a rate of 225 GB/s at a cost of 0.000014 USD/GB.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [38] Meiser, L. C. et al. Reading and writing digital data in DNA. Nat. Protoc. 15, 86–101 (2020). source

      Not checked in the free run.

  • May combine two claims.

    Interpretationreasoning

    Accounting for losses due to strict sequence filtering and after Von Neumann correction, a full NovaSeq 6000 run could generate true random numbers at a rate of about 300 kB/s at a cost of 600 USD/GB.

    “Accounting for losses due to strict sequence filtering, and after Von Neumann correction, a full NovaSeq 6000 run could generate true random numbers at a rate of ca. 300 kB/s at a cost of 600 USD/GB.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationreasoning

    Taking advantage of the stochastic properties of chemical reactions, true random numbers can be generated from DNA synthesis, offering a viable alternative for large volumes of randomness.

    “In this work, we have taken advantage of the stochastic properties of chemical reactions, generating true random numbers from DNA synthesis, offering a viable alternative for large volumes of randomness.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: The chemical synthesis of DNA involves intrinsically stochastic processes that can serve as a source of true randomness.

  • May combine two claims.

    Premiseown data

    Microsynth informed the authors that this amount of DNA is manufactured by fully automated machines within 8.75 h and can be obtained commercially for about USD 100.

    “Microsynth informed us that this amount of DNA is manufactured by fully automated machines within 8.75 h, and can be obtained commercially for a price of ca. USD 100.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationown data

    De-biased sets of DNA-generated bit streams are random, with all NIST tests passed.

    “Using NIST’s statistical test suite for evaluation of randomness, we find that de-biased sets of DNA-generated bit streams are random, with all tests passed.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Premisecitation

    Illumina technology is fully scalable, and high-throughput systems such as Illumina NovaSeq 6000 allow a throughput of up to 20 billion sequence reads within 36 h for a price of about USD 22,000.

    “Illumina technology is fully scalable and high throughput systems (such as Illumina NovaSeq 6000) allow a throughput of up to 20 billion sequence reads within 36 h43 for a price of ~USD 22,00044.”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites [43] Scalability for sequencing like never before. NovaSeq System Specifications https://emea.illumina.com/systems/sequencing-platforms/novaseq/specifications.html (2020).

      Not checked in the free run.

    • cites [44] Johnson, C. D. Genomics and Bioinformatics Service https://www.txgen.tamu.edu/new-novaseq-pricing-25feb2019/ (2019).

      Not checked in the free run.

  • May combine two claims.

    Interpretationreasoning

    The possible entropy of a random 60-mer is 4^60 = 10^36, which is significantly larger than the number of sequences in the pool, implying that every strand synthesized may be expected to be unique.

    “The possible entropy of a random 60-mer is 460 = 1036, and thus is significantly larger than the number of sequences in the pool. This implies that it may be expected that every strand synthesized is unique.”
    Quote in the paperSubstantial claim
30 of 30

  • The paper relies on this without support.

    Coupling efficiency differs for each building block and depends on variables such as utilization period of synthesis reagents or protecting groups.

    “In addition, the coupling efficiency differs for each building block, and is dependent on variables such as the utilization period of synthesis reagents by the manufacturers or the protecting groups attached to each building block.”

  • The paper relies on this without support.

    Oxidation can lead to G–T transversion, whereby G is chemically altered such that during DNA replication it can be exchanged for T.

    “Oxidation can lead to a phenomenon called G–T transversion, whereby the base G gets chemically altered such that during DNA replication steps, it can be exchanged for a base T.”

  • The paper relies on this without support.

    No studies of sequencing error trends have shown effects of nature and magnitude as seen in Fig. 3a–c.

    “To our best knowledge, no studies of sequencing error trends have shown effects of nature and magnitude as we see in Fig. 3a–c.”

  • The paper relies on this without support.

    Synthetic production of DNA is a stochastic chemical process.

    “The synthetic production of DNA is a stochastic chemical process with the advantage that the individual molecules in the synthesized DNA sequence can easily be identified and analyzed by next generation sequencing (NGS) technologies.”

  • The paper relies on this without support.

    The volumes of individual building blocks are not controlled to the nearest microliter.

    “Microsynth informed us in a discussion that the volumes of the individual building blocks are not controlled to the nearest microliter.”

  • The paper relies on this without support.

    The chemical synthesis of DNA involves intrinsically stochastic processes that can serve as a source of true randomness.

    “using the intrinsically stochastic processes of a chemical reaction”
    “making use of the stochastic process of chemical product formation”

  • The authors state this as an assumption.

    There is no bias in the data for the theoretical entropy calculation of 28 PB.

    “if there is no bias in the data”
7 of 7

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?

  • More advanced applications, such as random number generation or arithmetic operations, prove the versatility of DNA as an information medium.

    “More advanced applications, for example, random number generation or the possibility to perform arithmetic operations, prove the versatility of DNA as an information medium.[11,12]”

    From Integrating DNA Encapsulates and Digital Microfluidics for Automated Data Storage in DNA (2022) · cites this paper as [11]

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

  • NIST pass-rate claim conflicts with Table 1

    Minor

    The text says every test had a pass rate greater than 54/56, but Table 1 reports exactly 54/56 for the longest-run-of-ones test.

  • Two throughput rates are given for the Intel RNG

    Minor

    The introduction gives the Intel RNG a throughput of 500 MB/s, while Table 2 lists the Intel digital random number generator at 800 MB/s.

2 of 2

Cites

Flagged first, then by citations

1 of the 49 references couldn’t be matched to a known paper.

Research works printed but not linked: 2

  1. Schneider, B. Applied Cryptography (John Wiley and Son, 1996).
  2. von Neumann, J. & Forsythe, G. E. Var. Tech. Use. Connect. Random Digits Natl Bur. Stand. Appl. Math. Ser. 38, 36–38 (1951).

Cited by

Sorted by citations

Related

Authors
Meiser, L. C., Koch, J., Antkowiak, P. L., Stark, W. J., Heckel, R., & Grass, R. N.
Affiliations
  • ETH Zurich (CH)
  • Technical University of Munich (DE)
Corresponding author
Grass, Robert N.
ORCID
Venue
Nature Communications
Publisher
Springer Science and Business Media LLC
Published
2020-11-18
Volume
11
Issue
1
Pages
5869-5869
ISSN
2041-1723 (ISSN-L)
DOI
10.1038/s41467-020-19757-y
PMC
PMC7675991
OpenAlex
W3103867685
Semantic Scholar
617719561a8e8c13041997d00eb916aa322905e8
Cited by
45
References
49
Citations by year
2026: 6 · 2025: 7 · 2024: 11 · 2023: 5 · 2022: 8 · 2021: 7 · 2020: 1
FWCI
1.70
Type
journal article
Language
en
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)
  • mediaTUM – the media and publications repository of the Technical University Munich (Technical University Munich) (submittedVersion)
  • Repository for Publications and Research Data (ETH Zurich) (submittedVersion, cc-by)
  • ETH Zürich Research Collection (cc-by)
Funders
Microsoft
Topics
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › DNA and Biological Computing
  • Physical Sciences › Computer Science › Artificial Intelligence › Quantum Information and Cryptography
  • Physical Sciences › Computer Science › Computational Theory and Mathematics › Computability, Logic, AI Algorithms
Keywords
true random number generation, DNA synthesis, DNA sequencing, von Neumann extractor, bias removal, NIST randomness tests, encryption security
MeSH terms
Algorithms, Base Sequence, DNA, Humans, 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.

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