Source-linked AI summary
A Rewritable, Random-Access DNA-Based Storage System
S. M. Hossein Tabatabaei Yazdi, Yongbo Yuan, Jian Ma, Huimin Zhao, Olgica Milenkovic
TL;DR
Existing DNA-storage methods lacked precise partial access and supported read-only storage, motivating a rewritable random-access architecture. The paper combines mutually uncorrelated address sequences with constrained coding and DNA-editing methods, and demonstrates the approach on university texts with high estimated storage density. The results support DNA as a medium for both archival and rewritable storage applications.
Problem
Existing DNA-storage methods require reconstructing the whole text to retrieve small data fragments and support read-only storage.
Method
The architecture uses mutually uncorrelated address sequences, constrained coding, and DNA editing to support selective access and rewriting.
Results
4.9 × 10^20 B/g was estimated for the encoded information, and four access and rewriting experiments completed without readout errors.
Takeaways & Limitations
The demonstrated architecture supports DNA storage with random access and content rewriting alongside high storage density.
Takeaways & Limitations
At larger scales, substitution errors may accumulate across rewrite cycles and require more sophisticated coding than prefix matching.
Abstract
from arXiv · showhide
We describe the first DNA-based storage architecture that enables random access to data blocks and rewriting of information stored at arbitrary locations within the blocks. The newly developed architecture overcomes drawbacks of existing read-only methods that require decoding the whole file in order to read one data fragment. Our system is based on new constrained coding techniques and accompanying DNA editing methods that ensure data reliability, specificity and sensitivity of access, and at the same time provide exceptionally high data storage capacity. As a proof of concept, we encoded parts of the Wikipedia pages of six universities in the USA, and selected and edited parts of the text written in DNA corresponding to three of these schools. The results suggest that DNA is a versatile media suitable for both ultrahigh density archival and rewritable storage applications.
1 Results
The architecture stores 1000-bps DNA blocks with unique addresses for selective access, then uses PCR, sequencing, and DNA-editing methods to retrieve or rewrite content. Experiments demonstrated access and rewriting on jointly stored university texts.
- Storage and access: Unique addresses enable selective identification and amplification of individual DNA blocks for random access.Each 1000-bps block is tagged at both ends with specially designed address sequences.
- Storage and access: 1000-bps blocks balance storage overhead against synthesis cost, with 960 bases available for encoded content.The remaining 960 bases are divided into twelve 80-bps sub-blocks, each encoding six text words.
- Random access experiments: Three selected access queries identified, amplified, sequenced, and decoded the blocks containing the requested information.Primers corresponding to unique addresses were used for PCR selection and amplification before Sanger sequencing.
- Rewriting experiments: Short substrings were rewritten by editing their host 1000-bps blocks, whereas longer regions were replaced by synthesizing new uniquely primed sequences.The study used DNA editing for relatively short modifications and new sequence synthesis when rewrites exceeded several hundreds of bases.
- Rewriting experiments: 32 linear 1000-bps fragments were pooled for experiments selecting and amplifying one or three sequences.Amplification was verified by gel electrophoresis and by Sanger sequencing randomly sampled sequences from the pools.
- Storage density: 4.9 × 10^20 B/g was the estimated storage density for a 17 KB ASCII file encoded in 32,000 bps.The estimate reflects the nucleotide mass and encoded file size reported by the authors.
2 Methods
The architecture combines constrained DNA address sequences with prefix-synchronized encoding to support selective access, error control, and rewriting. Address construction uses balanced prefixes, Hamming separation, mutual uncorrelatedness, and folding avoidance, while experiments and coding results support reliable decoding and scalable addressing.
- Address Design: Address sequences constrain GC content, Hamming distance, mutual correlation, and secondary structures to improve stability, specificity, and amplification reliability.The construction uses approximately 50% GC content, large pairwise distance, uncorrelated prefixes and suffixes, and no folding structures.
- Experimental Validation: Three selected and amplified 1000-bp blocks were sequenced and rewritten with 100% accuracy.Gel electrophoresis confirmed tightly concentrated sequence lengths around 1000 bps, while Sanger sequencing verified the rewritten bases.
- Balanced Codes and Running Digital Sums: BRDS-derived codewords are mapped to DNA sequences with approximately balanced GC content across prefixes while preserving Hamming distance.The mapping translates binary codewords into DNA bases and supports the C1-C2 address constraints.
- Sequence Correlation: Mutually uncorrelated address sequences prevent undesired cross-hybridization and support selective retrieval without accidental selection of other blocks.The paper proves exponentially many mutually uncorrelated q = 4 sequences, with a constructive lower bound that preserves Hamming distance.
- Prefix-Synchronized DNA Codes: Prefix-synchronized encoding produces uniquely decodable strings and provides error detection with limited error correction while avoiding address patterns.The algorithm is uniquely decodable for 0 ≤ x < G_n,ℓ, and experiments showed modified prefixes could be uniquely mapped back to their originals.
3 Discussion
The proposed DNA storage architecture enables accurate random access and cost-efficient rewriting, demonstrated through four access and rewriting experiments without readout errors.
- The architecture combines new coding schemes with classical random-access codes to support accurate random access and cost-efficient rewriting.
- Unique addresses are prohibited from appearing elsewhere in encoded information, preventing undesirable cross-hybridization during selection and amplification.
- Four access and rewriting experiments produced no readout errors, according to post-selection and rewriting Sanger sequencing.
- High synthesis costs limited the prototype’s scope and size because long DNA blocks are expensive.
- The authors predict that synthesis costs will decrease rapidly in the very near future.
Supplementary Information
The supplementary information covers the paper’s encoding example, theoretical proofs, address sequences, procedures, experiments, and hybrid DNA-classical storage.
- The supplementary information begins with a working example encoding Wikipedia entries.
- It includes proofs of theorems supporting the proposed system.
- It describes the construction and properties of address sequences.
- It presents an example of the encoding and decoding procedure.
- It reports experimental synthesis, access, and rewriting of DNA storage sequences.
- It discusses hybrid DNA-based and classical storage.
1 Encoding Wikipedia entries: A Working Example
The working example encodes introductory Wikipedia sections for six universities using word-based, prefix-synchronized coding. Compared with uncompressed ASCII, it achieves an almost 1.7-fold improvement in description length, while requiring a larger dictionary.
- The files contain 17 KB of introductory Wikipedia text covering Berkeley, Harvard, MIT, Princeton, Stanford, and the University of Illinois Urbana-Champaign.The text contains 1,933 words, including 842 distinct words.
- The encoding groups six words into fragments and combines 12 fragments for prefix-synchronized encoding, producing 27 DNA blocks of length 1000 bps.
- Uncompressed ASCII represents the 12,874 characters using 90,118 bits at seven bits per character.The calculation is 12874 × 7 = 90118 bits.
- Almost 1.7-fold improvement in description length is obtained with prefix-synchronized codes compared with ASCII encoding.The word-based approach requires roughly 70-times larger dictionaries, but only one dictionary copy is needed.
- Fixed-length encoding doubles the word-encoding representation to 24 bits to avoid catastrophic error propagation.An extra bit also prevents very small integers from producing long runs of the first symbol in an address.
2 Proofs of Theorems
The proofs establish bounds and structural properties for mutually uncorrelated sequences and prefix-synchronized codes, supporting reliable avoidance of address patterns and unique decoding.
- u(n) denotes the largest possible size of a set of mutually uncorrelated words of length n.
- A constructive recursive procedure gives u(n) > 4 · (1.31)^n while preserving normalized minimum Hamming distances.
- Theorem 6 counts 4-ary strings that avoid a set of mutually uncorrelated address sequences through the generating function F(z).
- The number of sequences avoiding mutually uncorrelated sequences grows roughly as 4^n because the dominant pole of the generating function is close to 4.
- Prefix-synchronized encoding is uniquely decodable because its prefix-free components identify the relevant parameters recursively.
3 Address Sequences
The address sequences are designed as mutually uncorrelated, balanced, and well-separated DNA strings that support selective access while avoiding problematic sequence structures.
- The address sequences have length 20, 50% GC content, mutual uncorrelatedness, and pairwise Hamming distance exactly 10.They were also checked for the absence of secondary structures at room temperature.
- The experiment used unique address pairs flanking the two ends of each 1000-bp data block.
- Only the left-hand addresses were used for prefix-synchronized coding.
- Interleaved {G, C} and {A, T} bases in the left addresses provide GC balancing across address prefixes.
4 Encoding and Decoding Example
The example demonstrates encoding and decoding with the self-uncorrelated address P = AGCTG, including the computation of coding values and recursive recovery of the encoded input.
- The example uses the self-uncorrelated address string P = AGCTG.
- CodePSC(P, 8, 550) produces an encoded output that is decoded using DecodePSC(P, X).
- The decoding example recursively processes X = CCAAATCT by identifying its prefix contribution and continuing on the remaining substring.
5 Experimental Synthesis, Access and Rewrite of DNA Sequences
The experiments encoded university Wikipedia content into 1000-bps DNA blocks, selected requested blocks from a mixture, and demonstrated targeted rewriting using gBlock and OE-PCR methods.
- Experimental design: 27 1000-bps sequences encoded information from six university Wikipedia pages, with 26 synthesized after one sequence was rejected for secondary-structure complexity.The corresponding address primers were also synthesized.
- gBlock editing: The gBlock editing design synthesized edit-containing fragments and PCR-amplified the remaining sequence, using at least 30-bp homology for one-pot OE-PCR assembly.The described gBlock fragments ranged from 177 to 588 bps.
- OE-PCR editing: OE-PCR editing used general primers of at most 60 bps, with short edits encoded as primer overhangs before assembling the full 1000-bps rewrite.The final PCR reaction used three PCR products as templates.
- Selection and access: Selection from a mixture isolated the requested B1, B2, and B3 sequences as correct-length 1000-bps products, confirmed by subsequent sequencing.The mixture contained all 27 linear fragments.
- Method boundary: When a gBlock exceeded 500 bps, direct resynthesis was less costly than gBlock rewriting, so the gBlock method was not used.This establishes a cost boundary for the editing approach.
5.2 B2 mutation B2-M synthesis
B2 rewriting combined a 177-bp gBlock containing the edit with PCR-amplified sequence, then used OE-PCR to assemble and verify the complete 1000-bps product.
- B2 mutation synthesis: A 177-bp gBlock containing the entire B2 edited region was synthesized, while another B2 portion was PCR amplified from the original sequence.The gBlock and PCR product were combined for assembly.
- B2 mutation synthesis: OE-PCR combined the B2 gBlock and PCR products using five primerless cycles followed by 30 cycles with B2 primers.The reaction volume was 50 ul.
- B2 mutation synthesis: Gel isolation produced the correct 1000-bps B2-M band after OE-PCR assembly.The product was deposited on a gel substrate before band recovery.
- B2 mutation synthesis: Two primer pairs separately amplified the first and second parts of B2-M, using B2 as the template.The resulting PCR products are shown in Fig. S5.
5.3 B3 mutation B3-M synthesis
B3 rewriting assembled two edited 560-bp gBlocks with a 60-bp overlap, then used OE-PCR to produce a correctly sized 1000-bps product.
- B3 mutation synthesis: Two 560-bp gBlocks containing the two B3 mutation regions were synthesized with a 60-bp overlap.The overlapping fragments enabled assembly of the full edited sequence.
- B3 mutation synthesis: OE-PCR combined the two B3 gBlocks using five primerless cycles and 30 cycles with B3 primers.The reaction volume was 50 ul.
- B3 mutation synthesis: The B3 OE-PCR product yielded a single correctly sized 1000-bps band.The product was recovered from a gel substrate.
- B3 mutation synthesis: Three PCR products were generated from B3 as template for the B3 rewrite assembly.The products are shown in Fig. S8.
- B3 mutation synthesis: The final B3 OE-PCR reaction used the three PCR products and B3 primers to obtain one correct-size 1000-bps band.The reaction used five cycles without primers and 30 subsequent cycles with primers.
6 Hybrid DNA-Based and Classical Storage
The authors identify sequencing errors and scaling limits, then motivate a hybrid DNA–classical storage scheme for managing larger or repeatedly rewritten systems.
- Observed errors: Two erroneous symbols appeared in one strand during small-scale Sanger sequencing and were corrected using prefix matching.The correction succeeded in the reported small-scale experiment.
- Scaling limitation: In systems involving millions of blocks, sequencing errors may not be correctable by prefix matching alone.The authors identify this as a possible large-scale problem.
- Scaling limitation: Substitution errors may accumulate across rewrite cycles, while added parity checks can violate prefix properties, disturb GC balance, and require updating after each rewrite.The passage states that more sophisticated coding schemes may therefore be needed.