Section: Overall Objectives

Genomic data processing

The main goal of the GenScale project is to develop scalable methods, tools, and software for processing genomic data. Our research is motivated by the fast development of next-generation sequencing (NGS) technologies that provide very challenging problems both in terms of bioinformatics and computer sciences. As a matter of fact, the last sequencing machines generate Tera bytes of DNA sequences from which time-consuming processes must be applied to extract useful and pertinent information.

Today, a large number of biological questions can be investigated using genomic data. DNA is extracted from one or several living organisms, sequenced with high throughtput sequencing machines, then analyzed with bioinformatics pipelines. Such pipelines are generally made of several steps. The first step performs basic operations such as quality control and data cleaning. The next steps operate more complicated tasks such as genome assembly, variant discovery (SNP, structural variations), automatic annotation, sequence comparison, etc. The final steps, based on more comprehensive data extracted from the previous ones, go toward interpretation, generally by adding different semantic information, or by performing high-level processing on these pre-processed data.

GenScale expertise relies mostly on the first and second steps. The challenge is to develop scalable algorithms able to devour the daily DNA flow that tends to congest the bioinformatics computing centers. To achieve this goal, our strategy is to work both on space and time scalability aspects. Space scalability is correlated to the design of optimized and low memory footprint data structures able to capture all useful information contained in sequencing datasets. The idea is that hundreds of Giga bytes of raw data absolutely need to be represented in a very concise way in order to completely fit into a computer memory. Time scalability means that the execution of the algorithms must be as short as possible or, at least, must last a reasonable amount of time. In that case, conventional algorithms that were working on rather small datasets must be revisited to scale on today NGS data. Parallelism is a complementary technique for increasing scalability.

GenScale research is then organized along three main axes:

      $-$ Axis 1: Data structures

      $-$ Axis 2: Algorithms

      $-$ Axis 3: Parallelism

The first axis aims at developing advanced data structures dedicated to sequencing data. Based on these objects, the second axis provides low memory footprint algorithms for a large panel of usual NGS tools. Fast execution time is improved by the third axis. The combination of these three components allows efficient and scalable algorithms to be designed.