Compressed Gk Arrays for K-mer Counting

Compressed Gk Arrays for K-mer Counting

Edit. Before we proceed with the commentary, Rayan suggested

Here is a really nice paper that will push the state of the art in genome assembly (h/t: Rayan Chikhi).


Edit. Before we proceed with the commentary, Rayan suggested that the title is too narrow.

You should change the title: the main purpose of compressed gk arrays is

not kmer counting.

It is a bit misleading that they compare against jellyfish. In fact,

they’re comparing against the jellyfish hash table (that is used as a k-mer


The purpose of compressed gk arrays is to make queries on reads, it’s an



Paper is available here -

Scalable and Versatile k-mer Indexing for High-Throughput Sequencing Data

Philippe et al. (2011) proposed a data structure called Gk ar- rays for indexing and querying large collections of high-throughput sequencing data in main-memory. The data structure supports versa-tile queries for counting, locating, and analysing the coverage pro le of k-mers in short-read data. The main drawback of the Gk arrays is its

space-consumption, which can easily reach tens of gigabytes of main-memory even for moderate size inputs. We propose a compressed variant of Gk arrays that supports the same set of queries, but in both near-optimal time and space. In practice, the compressed Gk arrays scale up to much larger inputs with highly competitive query times compared to its non-compressed predecessor. The main applications include variant calling, error correction, coverage pro ling, and sequence assembly.

An earlier paper from the same group got published in BMC Bioinformatics.

Querying large read collections in main memory: a versatile data structure

How do Gk arrays work?

Gk arrays index all k-mers occurring within each read of the collection1 using a modi ed sux array and complementary tables [21]. It takes advantage from the fact that reads are often compared against themselves and that queried k-mers are taken from a read and can be given by a starting position rather than in extenso. Gk arrays o ers seven types of locate and counting queries: either for getting the read identi ers in which a k-mer occurs (Q1/Q2), occurs at most once (Q5/Q6), and the occurrence positions with (Q7) or without this restriction (Q3/Q4). Table 1 gives an overview of the queries and theoretical properties of the data structure. Gk

arrays uses a space proportional to the length of the read collection. Hence, indexing for instance a metagenomics dataset will exhaust the main memory of most computers. Gk arrays are also limited to queries on a single k value.

Written by M. //