Informational and linguistic analysis of large genomic sequence collections via efficient Hadoop cluster algorithms.

Motivation: Information theoretic and compositional/linguistic analysis of genomes have a central role in bioinformatics, even more so since the associated methodologies are becoming very valuable also for epigenomic and meta-genomic studies. The kernel of those methods is based on the collection of k-mer statistics, i.e. how many times each k-mer in {A,C,G,T}k occurs in a DNA sequence. Although this problem is computationally very simple and efficiently solvable on a conventional computer, the sheer amount of data available now in applications demands to resort to parallel and distributed computing. Indeed, those type of algorithms have been developed to collect k-mer statistics in the realm of genome assembly. However, they are so specialized to this domain that they do not extend easily to the computation of informational and linguistic indices, concurrently on sets of genomes. Results: Following the well-established approach in many disciplines, and with a growing success also in bioinformatics, to resort to MapReduce and Hadoop to deal with 'Big Data' problems, we present KCH, the first set of MapReduce algorithms able to perform concurrently informational and linguistic analysis of large collections of genomic sequences on a Hadoop cluster. The benchmarking of KCH that we provide indicates that it is quite effective and versatile. It is also competitive with respect to the parallel and distributed algorithms highly specialized to k-mer statistics collection for genome assembly problems. In conclusion, KCH is a much needed addition to the growing number of algorithms and tools that use MapReduce for bioinformatics core applications. Availability and implementation: The software, including instructions for running it over Amazon AWS, as well as the datasets are available at http://www.di-srv.unisa.it/KCH. Contact: umberto.ferraro@uniroma1.it. Supplementary information: Supplementary data are available at Bioinformatics online.

FASTdoop: a versatile and efficient library for the input of FASTA and FASTQ files for MapReduce Hadoop bioinformatics applications.

SUMMARY: MapReduce Hadoop bioinformatics applications require the availability of special-purpose routines to manage the input of sequence files. Unfortunately, the Hadoop framework does not provide any built-in support for the most popular sequence file formats like FASTA or BAM. Moreover, the development of these routines is not easy, both because of the diversity of these formats and the need for managing efficiently sequence datasets that may count up to billions of characters. We present FASTdoop, a generic Hadoop library for the management of FASTA and FASTQ files. We show that, with respect to analogous input management routines that have appeared in the Literature, it offers versatility and efficiency. That is, it can handle collections of reads, with or without quality scores, as well as long genomic sequences while the existing routines concentrate mainly on NGS sequence data. Moreover, in the domain where a comparison is possible, the routines proposed here are faster than the available ones. In conclusion, FASTdoop is a much needed addition to Hadoop-BAM. AVAILABILITY AND IMPLEMENTATION: The software and the datasets are available at http://www.di.unisa.it/FASTdoop/ . CONTACT: umberto.ferraro@uniroma1.it. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.