Supplementary material for The Human Genome Contracts Again

Genome compression has been the subject of multiple studies in the past several years. In general, studies in this field can be grouped into two main categories: reference-based compression and non-reference based compression. Methods which do not use reference genomes exploit the repetitive nature of genomic sequences, and compress the input data by using variations over general-purpose compression schemes, such as LZ77 [Lempel & Ziv 1977], which are modified to account for features specific to genomic data, such as inversions, palindromes, and approximate repeats. There are many studies in this category, including [Grumbach & Tahi 1994, Powell et al 1998, Gusev et al. 1999, Apostolico & Lonardi 2000, Matsumoto et al. 2000, Adjeroh et al. 2002, Chen et al. 2002, Tabus et al. 2003, Manzini & Rastero 2004, Behzadi & Le Fessant 2005, Korodi & Tabus 2005, Srinivasa et al. 2006, Cao et al. 2007, Korodi & Tabus 2007, Pinho et al. 2011, Kreft & Navarro 2011]. For a review, see [Giancarlo et al. 2009]. The best compression obtained with such approaches results in about 600MB per genome (see [Pinho et al. 2011] for comparison of several methods). Methods in the second category (reference-based) obtain much better compression by utilizing the fact that the genomes of different individuals are more than 99.8% identical and coding just the differences between the input genome and a reference genome. Studies in this category include [Christley et al. 2009, Brandon et al. 2009, Deorowics & Grabowski 2011b, Kuruppu et al. 2011, Wang & Zhang 2011, Hsi-Yang Fritz et al. 2011, Pinho et al. 2012, Chern et al. 2012]. The different methods differ in the way the reference sequence is selected or computed from a set of genomes, and the way the difference map is computed and compressed (usually using LZ-inspired schemes). Reference-based methods can reportedly reduce the size of the compressed genome to 3.1MB-19MB. In general, better results may be obtained when multiple sequences are compressed simultaneously, or when multiple reference genomes are used (note that two tasks are closely related), as discussed in [Mantaci et al. 2005, Mäkinen et al. 2010, Kuruppu et al. 2011, Deorowics & Grabowski 2011b, Kuruppu et al. 2012]. The best results we are aware of among schemes that simultaneously compress multiple genomes are reported in [Deorowics & Grabowski 2011b], who compress 70 complete human genomes to an average size of 3.1 MB per genome. The best compression using a single reference genome thus far was reported in [Christley et al. 2009], who compressed James Watson’s genome to 4MB, by using dbSNP to represent more efficiently known SNPs in the difference map. Another class of relevant studies focus on compression of short reads [Tembe at al. 2010, Deorowics & Grabowski 2011a, Hsi-Yang Fritz et al. 2011]. One of these schemes [Hsi-Yang Fritz et al. 2011] uses a reference genome against