Optimized relative Lempel-Ziv compression of genomes

  • Authors:

  • Shanika Kuruppu;Simon J. Puglisi;Justin Zobel

  • Affiliations:

  • The University of Melbourne, Parkville, Victoria;RMIT University, Melbourne, Victoria;The University of Melbourne, Parkville, Victoria

  • Venue:
  • ACSC '11 Proceedings of the Thirty-Fourth Australasian Computer Science Conference - Volume 113
  • Year:
  • 2011

Quantified Score

Hi-index 0.00

Visualization

Abstract

High-throughput sequencing technologies make it possible to rapidly acquire large numbers of individual genomes, which, for a given organism, vary only slightly from one to another. Such repetitive and large sequence collections are a unique challange for compression. In previous work we described the RLZ algorithm, which greedily parses each genome into factors, represented as position and length pairs, which identify the corresponding material in a reference genome. RLZ provides effective compression in a single pass over the collection, and the final compressed representation allows rapid random access to arbitrary substrings. In this paper we explore several improvements to the RLZ algorithm. We find that simple non-greedy parsings can significantly improve compression performance and discover a strong correlation between the starting positions of long factors and their positions in the reference. This property is computationally inexpensive to detect and can be exploited to improve compression by nearly 50% compared to the original RLZ encoding, while simultaneously providing faster decompression.