Fine-Grained Parallel Genomic Sequence Comparison

Abstract

Comparing DNA, RNA or protein sequences is a fundamental process in computational biology. The information deduced by processing genomic sequences remain the base of a large panel of bioinformatics activities such as genome assembly, gene annotation, phylogeny, prediction of 3D protein structures, meta-genomic analysis, etc. For almost two decades, the amounts of data have steadily increased, nearly doubling every 16-18 months. Hence, from gene level analyses, bioinformatics researches have moved to full genome analysis, leading to extremely large quantities of data to process. Furthermore, recent progresses in biotechnology, such as the next generation sequencing technology able to generate billions of genomic sequences in a single day, still strengthen the needs for fast and efficient solutions. Basically, genomic data, which are considered here, are DNA or protein sequences. A DNA sequence may be as simple as a single gene (a few thousands of nucleotides) or as complex as a full genome (three billions of nucleotides for the human genome). A protein sequence is shorter. It reflects the DNA to amino acids transcription of genes through the universal genetic code. Their lengths range from a few hundreds of amino acids to a few thousands of amino acids. The alphabet of a nucleotide sequence is composed of only 4 characters: A, C, G and T. The protein alphabet is larger and includes 20 amino acids. From a computational point of view, these data are seen as simple strings of characters. These sequences are stored in genomic databases. SWISS-PROT and TrEMBL (Apweiler et al., 2004), for example, are two well-known protein sequence databases containing respectively 466739 and 7695149 entries (May 2009). From the DNA size, GenBank (release 171, Apr. 2009) contain more than 100 millions of sequences, representing more than 100 billions of nucleotides (Benson et al., 2008). New releases are made every two months to include new data coming from worldwide research institutes. With the exponential growth of these databases, performing computation on this mass of data is every day a more and more challenging task. A lot of bioinformatics applications need to compare genomic sequences in their early processing steps. To illustrate our point, we briefly describe some of them in the next paragraphs. The goal is not to provide an exhaustive list, but to give, through some examples, an idea of the volume of data which are routinely processed. 14