Magnetic DNA random access memory with nanopore readouts and exponentially-scaled combinatorial addressing

Billy T. Lau, Shubham Chandak, Sharmili Roy, Kedar Tatwawadi, Mary Wootters, Tsachy Weissman, Hanlee P. JiScientific 2023 · DOI

Abstract

The storage of data in DNA typically involves encoding and synthesizing data into short oligonucleotides, followed by reading with a sequencing instrument. Major challenges include the molecular consumption of synthesized DNA, basecalling errors, and limitations with scaling up read operations for individual data elements. Addressing these challenges, we describe a DNA storage system called MDRAM (Magnetic DNA-based Random Access Memory) that enables repetitive and efficient readouts of targeted files with nanopore-based sequencing. By conjugating synthesized DNA to magnetic agarose beads, we enabled repeated data readouts while preserving the original DNA analyte and maintaining data readout quality. MDRAM utilizes an efficient convolutional coding scheme that leverages soft information in raw nanopore sequencing signals to achieve information reading costs comparable to Illumina sequencing despite higher error rates. Finally, we demonstrate a proof-of-concept DNA-based proto-filesystem that enables an exponentially-scalable data address space using only small numbers of targeting primers for assembly and readout.

Introduction

While DNA is an incredible medium for high-density, long-term data storage, it faces three major hurdles for practical implementation. First is molecular consumption: traditional sequencing often requires PCR amplification that consumes the original synthesized pool, meaning repeated access eventually "burns" the source material. Second is the nanopore error gap: Oxford Nanopore sequencing is portable but suffers from high error rates that typically require high "coverage" (reading the same strand many times), driving up costs. Finally, there is a scaling bottleneck: retrieving a specific file from a massive library traditionally requires a unique pair of primers for every file, which becomes logistically impossible as libraries grow to millions of files.

The MDRAM Framework

MDRAM addresses these issues through a multi-layered approach integrating physical, coding, and architectural innovations.

At the physical layer, the system uses TCO-Tz click chemistry to covalently bond synthesized DNA to magnetic agarose beads. This allows researchers to perform PCR, wash away the amplicons for sequencing, and keep the "master" DNA template safely attached to the beads for the next use, enabling true non-destructive random access.

The MDRAM system architecture: (A) The end-to-end workflow from raw data encoding to nanopore readout. (B) The concatenated coding scheme featuring Reed-Solomon outer codes and convolutional inner codes. (C) The molecular conjugation process using magnetic agarose beads.

At the coding layer, MDRAM implements a dual-code system. An outer Reed-Solomon code handles missing sequences (dropouts), while an inner convolutional code handles sequencing errors. Crucially, the decoder integrates with the Bonito basecaller to use "soft information"—the raw electrical current signals from the nanopore—rather than just the final text output. This allows the system to extract more information from noisy reads.

Architecturally, the system utilizes Combinatorial Barcode Addresses (CBAs). Instead of one primer per file, it uses a hierarchical scaffold of barcode subunits. By mixing and matching a small set of primers, it can address an exponentially larger number of unique locations, mimicking a computer's demultiplexer.

Experimental Validation and Results

The system was evaluated using two primary oligonucleotide pools, including Pool A, which contained 11.3 KB of data across 13 files (such as the Gettysburg Address and the UN Declaration of Human Rights).

MDRAM achieved a reading cost of 3–5 bases per bit, a 10x improvement over previous nanopore-based methods. This efficiency brings nanopore sequencing into the same league as high-accuracy Illumina sequencing. The system successfully decoded files with as little as 3x read coverage, whereas other frameworks often require 30x or more.

Validation of MDRAM performance: (A) qPCR analysis of standards and washes confirming high conjugation efficiency to magnetic beads. (B) Specificity of random access, showing the proportion of reads aligned to targeted vs. non-targeted files.

Targeting precision was equally robust; using the CBA filesystem, the researchers achieved a 17,600-fold enrichment of a targeted oligonucleotide compared to the rest of the pool. Furthermore, the system maintained high data quality across five sequential random-access experiments on the same batch of beads, proving the "RAM" capability of the technology.

Discussion and Future Outlook

MDRAM demonstrates that DNA storage can be transformed from a "write-once, read-once" medium into a reusable, scalable platform. By leveraging raw signal processing and magnetic bead conjugation, the system drastically reduces the cost and molecular overhead of data retrieval.

While synthesis costs and physical latency (hours to days for PCR and sequencing) remain challenges, the authors suggest that future work will focus on integrating this workflow with automated, robotic platforms. This could enable fully autonomous, high-density "cold" DNA data centers that compete with modern silicon-based cloud infrastructure.

Summary updated

Abstract

The storage of data in DNA typically involves encoding and synthesizing data into short oligonucleotides, followed by reading with a sequencing instrument. Major challenges include the molecular consumption of synthesized DNA, basecalling errors, and limitations with scaling up read operations for individual data elements. Addressing these challenges, we describe a DNA storage system called MDRAM (Magnetic DNA-based Random Access Memory) that enables repetitive and efficient readouts of targeted files with nanopore-based sequencing. By conjugating synthesized DNA to magnetic agarose beads, we enabled repeated data readouts while preserving the original DNA analyte and maintaining data readout quality. MDRAM utilizes an efficient convolutional coding scheme that leverages soft information in raw nanopore sequencing signals to achieve information reading costs comparable to Illumina sequencing despite higher error rates. Finally, we demonstrate a proof-of-concept DNA-based proto-filesystem that enables an exponentially-scalable data address space using only small numbers of targeting primers for assembly and readout.
BibTeX
@inproceedings{lau2023magnetic,
  author = {Billy T. Lau and Shubham Chandak and Sharmili Roy and Kedar Tatwawadi and Mary Wootters and Tsachy Weissman and Hanlee P. Ji},
  title = {{Magnetic DNA random access memory with nanopore readouts and exponentially-scaled combinatorial addressing}},
  booktitle = {Scientific},
  year = {2023},
  doi = {10.1038/s41598-023-29575-z},
}
Figure