Section: Scientific Foundations
Keywords : Evolution, genome organisation, genome dynamics, motifs, search, inference, phylogenetic reconstruction, probabilistic modelling, data analysis, text algorithms, tree algorithms, combinatorics, permutations, knowledge bases.
Comparative genomics
Participants : Sophie Abby, Vicente Acuña, Anne-Muriel Arigon, Bastien Boussau, Frédéric Boyer, Eric Coissac, Yves-Pol Deniélou, Vincent Daubin, Marc Deloger, Marília Dias Vieira Braga, Laurent Duret, Christian Gautier, Philippe Genoud, Jean-Francois Gout, Manolo Gouy, Laurent Guéguen, Claire Guillet, Daniel Kahn, Claire Lemaitre, Jean Lobry, Gabriel Marais, Dominique Mouchiroud, Sylvain Mousset, Anamaria Necsulea, Leonor Palmeira, Guy Perrière, François Rechenmann, Marie-France Sagot, Paulo Gustavo Soares da Fonseca, Eric Tannier, Raquel Tavares, Alain Viari, Danielle Ziébelin.
Comparative genomics may be seen as the analysis and comparison of genomes from different species in order to identify important genomic features (genes, promoter and other regulatory sequences, regions homogeneous for some characteristics such as composition etc.), study and understand the main evolutionary forces acting on such genomes, and analyse the general structure of the genomic landscape, how the different features relate to each other and may interact in some life processes.
Computationally speaking, comparative genomics requires expertise with knowledge representation, probabilistic modelling techniques, general data analysis and text algorithmic methods, phylogenetic reconstructions, and combinatorics. All such expertises are present in HELIX as reflected in past and current publications.
Computational analysis of the evolution of species and gene families
Evolution is the main characteristic of living systems. It creates biological diversity that results from the succession of two independent processes: one introducing mutations that allow the genetic information transmitted to a descendant to vary slightly in relation to the genetic information present in the parent organism, and another fixing the mutation, where the frequency of occurrence of a tiny fraction of the errors increases in the population until the errors become the norm.
The analysis of the origin and frequency of mutations, as well as the constraints on their fixation, in particular the effect of natural selection, underlies an important part of the field of molecular computational biology. It therefore appears in almost all research areas developed in the HELIX project.
The comparison of proteic or nucleic sequences allows the a priori reconstruction of the whole of the Tree of Life. However, the mathematical complexity of the processes involved requires methods for approximate estimation. Moreover, sequences are not the only source of information available for reconstructing phylogenetic trees. The order of the genes along a genome is undergoing progressive change and the comparison of the permutations observed offers another way of estimating evolutionary distances. The methodological problems encountered are mainly related to the estimation of such distances in terms of the number of elementary (and biologically meaningful) operations enabling one permutation to succeed another. Sophisticated algorithms are required to deal with the problem. Once phylogenetic trees have been constructed, other problems arise that concern their manipulation and interpretation. Currently, more than 6000 families of genes (having more than 4 specimens ) are known, and hence can be represented by more than 6000 different trees (HELIX also developed specialized databases to hold this kind of information). The management, comparison and update of these trees represents a challenging computational and mathematical problem.
Modelling and analysis of the spatial organisation and dynamics of genomes
Genomic sequences are characterized by strong biological and statistical heterogeneities in their composition and organisation. In fact, neighbouring genes along a genome often share multiple properties, whose nature is structural (size and number of introns), statistical (base and codon frequencies), and linked to evolutionary processes (substitution rates). In certain cases, such neighbouring structures have been interpreted in terms of biological processes. For instance, in bacteria the spatial organisation of genomes results in part from the mechanism of replication. Other local structures, however, still resist the discovery of a mechanism that could explain their generation and maintenance. The most characteristic example in vertebrates concerns isochores usually defined as regions that are homogeneous in terms of their G+C composition. The identification of isochores is essential for the annotation of sequences as it correlates with various other genomic features (base frequency, gene structure, nature of transposable elements). The analysis of the spatial structure of a genome requires the elaboration of correlation methods (non-parametric correlation determination along a neighbour graph and Markov processes) and of partitioning (or segmentation) techniques.
In the course of evolution, the spatial organisation of a genome undergoes several changes that are the result of biological processes also not yet fully understood, but which generate various types of modifications. Among these changes are permutations between closely located genes, inversion of whole segments, duplication, and other long-range displacements. It is therefore important to be able to define a permutation distance that is biologically meaningful in order to derive true evolutionary scenarios between species or to compare the rates of rearrangements observed in different genomic regions. The HELIX project has been particularly interested in elaborating an operational definition for the notion of synteny in bacteria and in eukaryotes (two completely different notions for the two kingdoms). The elaboration of these definitions, together with their precise mathematical characterizations require expertise both in biology and in computer science.
Motif search and inference
The term motif is quite general, referring to locally-conserved structures in biological entities. The latter may correspond to biological sequences and 3D structures, or to abstract representations of biological processes, such as evolutionary trees or graphs, and biochemical or genetic networks. When referring to sequences, the term motif must be understood in a broad sense, which covers binding sites in both nucleic and amino acid sequences, but also genes, CpG islands, transposable elements, retrotransposons, etc.
The occurrence of motifs in a sequence provides an indication of the function of the corresponding biological entity. Identifying motifs, whether using a model established from previously-obtained examples of a conserved structure or proceeding ab initio , represents therefore an important area of research in computational biology. Motif identification consists of two main parts: 1. feature identification, which aims at finding and precisely mapping the main features of a genome: protein or RNA-coding genes, DNA or RNA sequence or structure signals, satellites (tandem repeats) or transposable elements (dispersed repeats with a specific structure), regulatory regions, etc; 2. relational identification, the goal of which consists in finding relations existing among the features individually characterized in the first step. Such relations are diverse in nature. They may, for instance, concern the participation of various features in a cellular process, or their physical interaction.
Search and inference problems, whether they concern features or relations, are in fact the extremes of a continuum of problems that range from seeking for something well-known to trying to identify unknown objects. The main difficulty lies in the fact that features and the relations holding between them should in general be inferred together. However, the information that must be manipulated in this case (cooperative signals, operons, regulons, reaction pathways or molecular assemblies) is more complex than the initial genome data and thus requires a higher degree of abstraction, and more sophisticated algorithms or statistical approaches. Various search and inference methods have already been developed by HELIX. These include methods for DNA and protein sequence motifs inference, gene finding, satellites and repeats identification and RNA common substructure inference. More recent work concerns the definition of motifs in graphs representing, for instance, metabolic pathways. In the last year, work has started also on mixing information from various often quite heterogeneous sources to infer motifs. These include for now sequence information with information on gene expression coming from microarray experiments and information provided by the signals evolution imprints into genomes. The final objective is to be able to automatically infer whole cellular modules, that is in fact small or, in the longer term, larger-scale biological networks. This new topic provides a strong link between comparative and functional genomics, molecular biology seen at the linear level of a genome and networks.
