Promiscuous molecules for smarter file operations in DNA-based data storage

The accelerating pace of information generation necessitates innovative data storage solutions that overcome limitations in conventional media. DNA offers significant advantages due to its extremely high density, remarkable durability, and efficient resource use. Recent advancements in DNA synthesis and sequencing have enabled the creation of DNA-based data storage systems exceeding 1 GB, suggesting commercial viability in the near future. However, crucial challenges remain in organizing, accessing, and searching this densely packed information. Current DNA storage systems typically store data as numerous distinct DNA molecules, free-floating in close proximity. This necessitates an addressing system that functions efficiently in a complex molecular mixture, while avoiding the space constraints of a physical scaffold. Existing methods utilize specific DNA base-pair interactions with short address sequences for file access, but these interactions are not strictly all-or-none, limiting storage capacity and functionality. Furthermore, metadata-based search methods struggle to differentiate files with similar content. A system for previewing low-resolution versions of files before full access would greatly improve efficiency. The paper proposes leveraging "nonspecific" interactions – conventionally viewed as a hindrance – to enhance address space, storage capacity, and implement advanced functionalities such as File Preview, drawing inspiration from techniques used in DNA editing and in-storage search.

Prior DNA-based storage systems relied on specific DNA base-pair interactions with ~20-nucleotide address sequences for file access, using PCR-based amplification or hybridization-based separations. To minimize unwanted cross-interactions between addresses, these systems imposed strict similarity thresholds (e.g., Hamming distance), reducing storage capacity and limiting metadata incorporation. These limitations restrict the development of more sophisticated functionalities. This research builds upon existing work in DNA storage but directly addresses the shortcomings of previous approaches by exploring the potential of thermodynamically tunable nonspecific interactions for improved file access and organization. The authors cite previous work in DNA editing and in-storage search as inspiration for this new approach.

The study began with a theoretical and experimental investigation of factors affecting DNA-DNA interactions. A Monte Carlo simulation incorporating the NuPACK thermodynamic model showed that a Hamming distance greater than 10 was needed to minimize unwanted hybridizations, confirming experimental findings. However, this stringent criterion severely limits address space. The authors hypothesized that tunable nonspecific interactions could be exploited to access different subsets of DNA strands by altering environmental conditions (temperature, primer concentration). They experimentally validated this hypothesis, demonstrating that lower temperatures and higher primer concentrations increase nonspecific amplifications, while higher temperatures and lower concentrations decrease them. A competitive PCR system with two unique template strands having closely related addresses was used to further investigate this tunability. The researchers then explored the implementation of a "File Preview" function using this tunable promiscuity. They encoded individual files into distinct subsets of strands, allowing differential access to low-resolution previews or full-resolution images using the same primers under different PCR conditions. This was tested with four JPEG images, in both isolated conditions and in the presence of a 1.5 GB randomized background to simulate a realistic data storage system. The impact of various PCR parameters (temperature, primer and magnesium chloride concentration, potassium chloride concentration, and the presence of detergents) on amplification specificity/promiscuity was evaluated to optimize File Preview functionality. Error-prone PCR was used to create a noisy background equivalent to 1.5 GB of data, allowing for testing the robustness of the File Preview function in a realistic scenario. Next-generation sequencing and subsequent data analysis (including clustering with the Starcode algorithm) were employed to quantify amplification results and decode the images. The design of the JPEG encoding within DNA sequences is thoroughly described in Supplementary Figure 7 and explained further in the supplementary materials. The researchers developed both experimental and computational methods, including Monte Carlo simulations (to model hybridization likelihoods based on Hamming distance) and a novel python program to analyze the results.

The study demonstrated that PCR stringency is thermodynamically tunable, allowing control over nonspecific amplifications by manipulating temperature and primer concentration. This tunability was successfully applied to implement a File Preview function, allowing selective access to low-resolution previews or full-resolution images of JPEG files using the same primers under different PCR conditions. The File Preview function worked reliably in both isolated and high-background (1.5 GB of randomized data) conditions. The authors observed that the most significant factor affecting Preview tunability was annealing temperature, with lower temperatures favoring amplification of mismatched strands. Other parameters, such as primer and magnesium chloride concentrations, also contributed to fine-tuning the system. Notably, the accidental creation of strands with 5 HD binding sites revealed a clear distinction in access between intermediate and full file access conditions (a transition of just 1 HD). While the initial design intended for 0, 4, and 6 HD strands per file, problematic sequences within the data payload region (resulting in 5 HD binding sites) indicated the need for careful quality control measures during codeword combinations and careful primer design to avoid unintended interactions. The study demonstrated cost savings from this system, showing that for a database of 15 highly similar files, a 5% File Preview system would reduce the sequencing cost by 85.3% compared to a system requiring full sequencing of each file. Even with a 1% File Preview, sequencing costs were reduced by 91.7%. The analysis of Next-Generation Sequencing data, including clustering using the Starcode algorithm, enabled precise quantification of the relative abundance of different strands. A more detailed description of the process of decoding the NGS data is described in the Supplementary Materials.

The successful implementation of File Preview demonstrates that thermodynamically tunable promiscuity can provide valuable functionality in DNA data storage systems. This function significantly reduces the cost and time required for file searching and retrieval in databases with many similar files, making it highly practical. The ability to selectively access subsets of data opens up opportunities for other functions, such as differential encoding of frequently and infrequently accessed data or data deduplication. Although the presence of unintended 5 HD binding sites revealed a potential limitation, this can be mitigated through careful quality control during the encoding process. The potential for off-target interactions in the data payload regions under promiscuous conditions warrants further investigation. However, the results suggest that interactions within the same file's strands with higher HD addresses are more likely than interactions with undesired files.

This research successfully demonstrated a novel method for improving data access and organization in DNA-based data storage by leveraging tunable promiscuous interactions. The implemented File Preview function significantly reduces costs associated with searching and retrieving files, especially within databases containing numerous similar files. Future research could focus on refining the encoding process to eliminate unintended interactions, optimizing the design of access conditions, and expanding the application of this tunable promiscuity to other functionalities, such as data deduplication and prioritizing data based on frequency of access.

The study revealed limitations related to the potential for unintended interactions within the data payload region, especially when using promiscuous conditions. Careful quality control during the encoding process is crucial to minimize this issue. Further investigation is needed to fully understand the extent of off-target interactions in large-scale systems and to optimize conditions to further reduce these interactions. The current balance of Preview versus full-access strands (5% File Preview) results in a tradeoff in terms of physical storage density, though this can be mitigated by altering the ratio of full file strands to preview strands, or by further optimization of PCR conditions.