By Kevin Davies
January 30 | A new report published in the open-access journal Bioinformatics describes a method of compressing a whole human genome sequence into a file sufficiently small to serve as a standard email attachment.
The work was published by Scott Christley, Yiming Lu, Chen Li and Xiaohui Xie at the University of California, Irvine.
The authors note in their short paper (available here) that the 1000 Genomes Project is generating on the order of 8 billion bases per day, or the equivalent of a new GenBank each week. The total amount of sequencing information generated, 60 trillion bases, represents “60 times more sequence than all the public DNA data that have been deposited over the past 25 years,” the authors write.
While programs such as gzip can reduce the size of a genome file considerably, Christley and colleagues opted to create a novel algorithm that would reduce the size of a single genome by several orders of magnitude. The trick is not to compress the entire individual genome from scratch, but to document all of the 1% or so of variations from the reference sequence.
The authors use four tiers of compression:
1) VINT (variable integers for positions) – stores the position for each variation, including SNPs and indels.
2) DELTA (delta positions) – stores the position value as a relative position from the previous site of variation
3) DBSNP (SNP mapping) – incorporates information on common SNPs from a database such as dbSNP
4) KMER (K-mer partitioning) – uses Huffman coding to encode the insertion sequence data.
Applying these forms of compression in addition to gzip, Christley and coworkers were able to compress Jim Watson’s genome sequence from a 3+ gigabyte file to 84 a mere 4.1 megabytes. At that size, the genome is small enough to be an email attachment.
The paper did not discuss the ease of reconstituting the ultra-compressed genome. Moreover, the technique requires the existence of a reference human genome and a reference SNP map, “but this cost is amortized over the total number of genomes,” the authors write.
Further reading: Christley, S. et al. Bioinformatics 25, 274-275 (2009). Doi: 10.1093/bioinformatics/btn582