How DNA could store all the world’s data

Abstract

The team used many short DNA strings to encode a 659-kB version of a book Church had co-authored. Part of each string was an address that specified how the pieces should be ordered after sequencing, with the remainder containing the data. A binary zero could be encoded by the bases adenine or cytosine, and a binary one could be represented by guanine or thymine. That flexibility helped the group to design sequences that avoided reading problems, which can occur with regions containing lots of guanine and cytosine, repeated sections, or stretches that bind to one another and make the strings fold up. They didn't have error correction in the strict sense, instead relying on the redundancy provided by having many copies of each individual string. Consequently, after sequencing the strings, Kosuri, Church and Gao found 22 errors — far too many for reliable data storage. At the EBI, meanwhile, Goldman, Birney and their colleagues were also using many strings of DNA to encode their 739-kb data store, which included an image, ASCII text, audio files and a PDF version of Watson and Crick's iconic paper on DNA's double-helix structure. To avoid repeat-ing bases and other sources of error, the EBI-led team used a more com-plex scheme (see 'Making memories'). One aspect involved encoding the data not as binary ones and zeroes, but in base three — the equivalent of zero, one and two. They then continuously rotated which DNA base represented each number, so as to avoid sequences that might cause prob-lems during reading. By using overlapping, 100-base-long strings that progressively shifted by 25 bases, the EBI scientists also ensured that there would be four versions of each 25-base segment for error-checking and comparison against each other.