Low cost DNA data storage using photolithographic synthesis and advanced information reconstruction and error correction

DNA's longevity and high information density make it an attractive archival storage medium. However, current DNA data storage systems are hampered by the high cost and slow speed of DNA synthesis. This research explores a shift from low-error, expensive synthesis to cheaper, potentially faster, high-error synthesis technologies. The core hypothesis is that advanced error correction algorithms can compensate for the higher error rates inherent in these cheaper methods. This approach aims to overcome the cost bottleneck in DNA data storage, making it a more viable alternative to traditional methods. The importance of this study lies in its potential to revolutionize long-term data archiving by significantly reducing costs while maintaining data integrity through robust error correction techniques. The work builds upon previous research in DNA storage, acknowledging the achievements of Goldman et al. and Church et al. in storing digital information in DNA, while proposing a more cost-effective approach by utilizing a high-error synthesis technology, coupled with sophisticated error correction techniques. The goal is to demonstrate the feasibility of this novel approach and to assess its potential for scalability and cost-effectiveness.

Early breakthroughs in DNA data storage, such as those by Goldman et al. and Church et al., established the principle of using synthesized DNA molecules to store digital information. These studies used high-fidelity, but expensive, DNA synthesis techniques. Recent advances in parallelized DNA synthesis and sequencing technologies, including portable sequencing devices, fueled further interest in DNA as a storage medium. Enzymatic methods for DNA synthesis were also investigated as an alternative to traditional phosphoramidite methods; however, these methods are not yet advanced enough to create highly variable DNA libraries. This paper builds upon this existing body of work by exploring the potential of lower-cost, higher-speed, and higher-error synthesis methods, specifically focusing on the feasibility of using photochemical synthesis with its associated challenges.

The researchers employed light-directed maskless array technology for DNA synthesis. This method, unlike the electrode array-based technology or material deposition methods used by commercial suppliers, promises greater scalability at lower cost. The system uses digital micromirror devices (DMDs), enabling rapid synthesis of arbitrary sequences with minimal hardware. Given the relatively low coupling yields (95-99%) of this technology, 60-nt long oligos were synthesized, shorter than those used in previous DNA storage applications. Sequence amplification sites were omitted to maximize information storage within the oligo length. The DMD device's full resolution was not used to optimize speed, leading to a final exposure resolution of 128x128, allowing for the synthesis of 16,384 oligos. To store data, a conservative 60% of nucleotides were allocated for error correction code redundancy, enabling storage of approximately 100 kB of data. The chosen data comprised 52 pages of Mozart's String Quartet "The Hunt" K458. The encoding scheme involved pseudo-randomization of data to avoid homopolymers and unbalanced GC contents, followed by Reed-Solomon coding (both inner and outer) and indexing to manage errors. After synthesis, oligos were cleaved from the solid support and prepared for Illumina sequencing using a ssDNA library preparation kit from Swift Biosciences, which allowed for the creation of dsDNA with adapter sequences. Illumina NextSeq sequencing yielded 30 million reads with an average length of 63 nt. The error correction pipeline included clustering the sequences using a locality-sensitive hashing-based algorithm, aligning sequences within clusters using MUSCLE, and extracting candidate sequences through weighted majority voting. The inner and outer decoders were then applied to recover the original data. To validate the error correction approach, Monte-Carlo simulations were performed to model the error introduction during synthesis and match the observed patterns of errors.

The photochemically synthesized DNA exhibited high error rates: 2.6% substitution, 6.2% deletion, and 5.7% insertion errors. Error rates increased drastically towards the 3' end of the sequences. Analysis of base frequencies revealed that these errors stemmed primarily from statistical nucleotide deletions during synthesis, coupled with a C-rich tail added during library preparation. Despite these high error rates, the developed error correction pipeline perfectly recovered the 100 kB of Mozart's sheet music. The pipeline's critical components were the locality-sensitive hashing-based clustering, multiple alignment, and weighted majority voting to extract candidate sequences. The outer code successfully corrected 3.4% erasures and 13.3% errors. Scaling experiments were conducted with larger files (323 kB and 1.3 MB), which showed increased error probabilities, requiring more redundancy for perfect recovery. Cost analysis revealed that the photolithographic method, even with the higher error rates, is significantly cheaper than existing DNA synthesis methods. The cost is estimated to be approximately 530 US$/MB at this stage and holds potential for further reduction to 1 US$/MB with optimization.

This study successfully demonstrates that high-quality DNA synthesis is not a prerequisite for DNA data storage. The use of a lower-cost, higher-error synthesis method coupled with a sophisticated error-correction pipeline makes DNA data storage more economically feasible. The results directly address the research question of whether cost-effective, high-error synthesis could be used for reliable DNA data storage. The perfect recovery of the Mozart sheet music showcases the effectiveness of the developed encoding and decoding pipeline in handling high error rates. The significantly lower cost compared to previous methods suggests a viable path towards large-scale DNA data archiving. The scalability testing reveals the importance of adjusting redundancy to accommodate higher error rates as data size increases. This work challenges the conventional focus on high-fidelity synthesis in DNA data storage, opening up possibilities for further cost reduction through technological advancements and optimization of the synthesis process.

This research presents a significant advancement in DNA data storage by demonstrating successful data recovery from high-error DNA synthesized via photolithography. The combination of this cost-effective synthesis method and an advanced error correction pipeline holds promise for making DNA-based data storage a viable, affordable alternative for long-term archiving. Future research should focus on further optimizing the synthesis process to reduce errors, enhancing the error correction pipeline to handle even higher error rates, and exploring the scalability of this approach for storing terabytes or petabytes of data. Cost reductions, particularly in reagent usage and synthesis parameters, are key areas for future improvements.

The current study focuses on relatively small data sets (up to 1.3 MB). Scaling to much larger datasets might require further optimization of the error correction techniques and increased redundancy. The error rates, while significantly higher than in previous studies, might still be too high for some applications. Further research is needed to address these challenges and to fully evaluate the long-term stability and reliability of DNA data storage using this approach.