Robust retrieval of data stored in DNA by de Bruijn graph-basedde novostrand assembly

Abstract

High density and long-term features make DNA data storage a potential media. However, DNA data channel is a unique channel with unavoidable ‘data reputations’ in the forms of multiple error-rich strand copies. This multi-copy feature cannot be well harnessed by available codec systems optimized for single-copy media. Furthermore, lacking an effective mechanism to handle base shift issues, these systems perform poorly with indels. Here, we report the efficient reconstruction of DNA strands from multiple error-rich sequences directly, utilizing a De Bruijn Graph-based Greedy Path Search (DBG-GPS) algorithm. DBG-GPS can take advantage of the multi-copy feature for efficient correction of indels as well as substitutions. As high as 10% of errors can be accurately corrected with a high coding rate of 96.8%. Accurate data recovery with low quality, deep error-prone PCR products proved the high robustness of DBG-GPS (314Kb, 12K oligos). Furthermore, DBG-GPS shows 50 times faster than the clustering and multiple alignment-based methods reported. The revealed linear decoding complexity makes DBG-GPS a suitable solution for large-scale data storage. DBG-GPS’s capacity with large data was verified by large-scale simulations (300 MB). A Python implementation of DBG-GPS is available at https://switch-codes.coding.net/public/switch-codes/DNA-Fountain-De-Bruijn-Decoding/git/files.