A digital twin for DNA data storage based on comprehensive quantification of errors and biases

Abstract

Archiving data in synthetic DNA offers unprecedented storage density and longevity. Handling and storage introduce errors and biases into DNA-based storage systems, necessitating the use of Error Correction Coding (ECC) which comes at the cost of added redundancy. However, insufficient data on these errors and biases, as well as a lack of modeling tools, limit data-driven ECC development and experimental design. In this study, we present a comprehensive characterisation of the error sources and biases present in the most common DNA data storage workflows, including commercial DNA synthesis, PCR, decay by accelerated aging, and sequencing-by-synthesis. Using the data from 40 sequencing experiments, we build a digital twin of the DNA data storage process, capable of simulating state-of-the-art workflows and reproducing their experimental results. We showcase the digital twin’s ability to replace experiments and rationalize the design of redundancy in two case studies, highlighting opportunities for tangible cost savings and data-driven ECC development.

Challenges for error-correction coding in DNA data storage: photolithographic synthesis and DNA decay

“…These error rates are in-line with the analysis by the original authors, 11,12 and in stark contrast to the error rates commonly observed for established commercial DNA synthesis, which usually lie below 0.02 nt -1 even in the worst case. 9,18 Interestingly, deletion errors dominate both in photolithographic and commercial DNA synthesis. 9,18,21 However, similarly to electrochemical synthesis in which a failure to deprotect the growing oligo causes a deletion, 22,23 part of the high error rates during photolithographic synthesis could be attributed to insufficient deprotection during the illumination step.…”

“…9,18 Interestingly, deletion errors dominate both in photolithographic and commercial DNA synthesis. 9,18,21 However, similarly to electrochemical synthesis in which a failure to deprotect the growing oligo causes a deletion, 22,23 part of the high error rates during photolithographic synthesis could be attributed to insufficient deprotection during the illumination step. The two aforementioned high-density syntheses by Antkowiak et al 12 did not show the same error distribution, as substitution errors were dominating instead (see File 2 and File 3 in Fig.…”

“…1,8 In doing so however, many error-correction codes are designed for low-error, low-bias scenarios involving highfidelity DNA synthesis and sequencing, without aging-induced DNA decay. 1,9,10 As a result, two important new additions to this state-of-the-art DNA data storage workflow currently lack established and optimized implementations of error-correction codes: photolithographic synthesis and DNA decay (see Fig. 1).…”

“…16 As this recovers only intact, full-length oligos -a small fraction of the available sequence information 17 -DNA decay is commonly limiting the effective storage density (i.e., code rate / physical coverage) by requiring high physical coverage. 9,10,18 Thus, achieving storage densities close to the theoretical limit 7 of 455 EB g -1 will require data recovery from low physical coverages rather than marginally improved code rates. To this end, the decoding of oligo fragments after DNA decay is a viable option, at the cost of increased complexity during decoding, 10,17,19 as shown by Song et al 19 and Meiser et al 17 (see Fig.…”

“…1) -and highlight the associated challenges for error-correction codes. To do so, we use analytical tools previously developed for DNA data storage 9 with existing sequencing data from literature. 11,12,17,19 In discussing the observed error patterns and biases, we identify opportunities for the optimization of experimental workflows and the development of error-correction codes.…”