Robust data storage in DNA by de Bruijn graph-based decoding

Abstract

Data storage in DNA, which store information in polymers, is a potential technology with high density and long-term features. However, the indels, strand rearrangements, and strand breaks that emerged during synthesis, amplification, sequencing, and storage of DNA molecules need to be handled. Here, we report a de Bruijn graph-based, greedy path search algorithm (DBG-GPS), which can efficiently handle all these issues by efficient reconstruction of the DNA strands. DBG-GPS achieves accurate data recovery with low-quality, deep error-prone PCR products, and accelerated aged DNA samples (solution, 70℃ for two weeks). The robustness of DBG-GPS was verified with 100 times of multiple retrievals using PCR products with massive unspecific amplifications. Moreover, DBG-GPS shows linear decoding complexity and more than 100 times faster than the multiple alignment-based methods, indicating a suitable solution for large-scale data storage.

A robust and efficient DNA storage architecture based on modulation encoding and decoding

“…0.3%). The methods by Song et al 16 and Yadiz et al 14 have a higher logical density than ours: 1.5 and 1.75, but can only correct errors up to 10% and 12%. In coding theory, the more logical redundancy is added, the more powerful the error correction ability.…”

“…However, other encodings have to take hash or convolution operations 15 , add RS/LDPC/CRC redundancy 8, 13 , or perform XOR randomization on the original binary stream before translating them into DNA sequences. Fountain code 22 and the de Brujin graph method by Song et.al 16 both need a filter process to discard binary streams containing illegal subsequences. Moreover, it should be noted that Yadiz et.al 14 and Xue et al 19 also adopted a modulation encoding which were essentially different from ours.…”

“…However, other encodings have to take hash or convolution operations 23 , add RS/LDPC/CRC redundancy 7,24 and might even perform XOR randomization on the original binary stream before translating them into DNA sequences. In addition, some encoding methods, such as the fountain code 21 and the de Brujin graph method by Song et.al 31 , need a filter process to discard binary streams containing illegal subsequences .…”

“…Compared to our method, the ones in Lenz et.al 24 , Antkowiak et.al 7 , and Press et.al 23 correct less errors (up to 18%, 14.5%, and 10%), and have lower logical densities(0.5, 0.8, and 0.5). The method by Song et.al 31 can correct up to 10% errors with a relatively high logical density of 1.5. To tolerate higher errors, these methods would have to add more logical redundancy, which will further lower their logical density.…”

“…15 proposed a special convolution code, HEDGES, which could use A* algorithm to search for the possible path, and other failure path were regarded as erasures to be corrected in the outer code. Song et al 16 constructed a de Bruijn graph (DBG) using multiple sequences and infer the correct sequence by a graph searching algorithm. For English text storage, our group constructed a 6-base code table for characters and the IDSs were corrected by distance-based code correction, multiple sequence voting, and spelling check 17, 18 .…”