Algorithmique de l'alignement structure-séquence d'ARN : une approche générale et paramétrée

Philippe Rinaudo 1, 2
2 AMIB - Algorithms and Models for Integrative Biology
LIX - Laboratoire d'informatique de l'École polytechnique [Palaiseau], LRI - Laboratoire de Recherche en Informatique, UP11 - Université Paris-Sud - Paris 11, Inria Saclay - Ile de France
Abstract : The alignment of biological macromolecules such as proteins, DNA or RNA is a biological and bio-informatics problematic which aims to reveal some of the mysteries of how cells works. The non-coding RNA are involved in the metabolism of all living beings. The two major issues concerning them are: the prediction of their structure to better understand their function and their detection in databases or genomes. One approach, the structure-sequence alignment of RNA, addresses these two issues. The work done during my thesis provides some constructive elements on this problem and led me to call the graph algorithmic for its resolution. The alignment problem is to align a structure of a first RNA with the sequence of a second RNA. The structure on the first RNA is represented as a graph or equivalently as an arc-annotated sequence and the sequence represents the nucleotide sequence of the second RNA.To solve this problem, we aim to compute a minimal cost alignment, according to a given cost function. So, this is an optimization problem, which turns out to be NP-hard.Accordingly, different works define several reduced structure classes for which they propose specific algorithms but with polynomial complexity. The work of my thesis unifies and generalizes previous approaches by the construction of a unique (not class specific) parameterized algorithm. Using this algorithm, it is possible to solve the problem of structure-sequence alignment for all possible instances, and as effectively as previous approaches in their respective field of resolution.This algorithm uses a technique from graph theory: the tree decomposition, that is to say, it transforms the given structure into a tree-decomposition and the decomposition is then aligned with the sequence. The alignment between a tree-decomposition and a sequence is done by dynamic programming. Its implementation requires a reformulation of the problem as well as a substantial modifications to the conventional use of dynamic programming for tree decompositions. This leads to an algorithm whose parameter is entirely related to the tree-decomposition.The construction of tree decompositions for which the alignment is the most effective is unfortunately a NP-Hard problem. Nevertheless, we have developed a heuristic construction of decompositions adapted to RNA structures. We then defined new structure classes which extend existing ones without degrading the complexity of the alignment but which can represent the majority of known structures containing many important elements that had not be taken into account previously (such as RNA tertiary motifs).The sequence-structure alignment problem attempts to answer the problem of prediction of structures and RNA research. However, the quality of the results obtained by its resolution depends on the cost function. During my PhD I started to define new cost functions adapted to the new structure classes by a machine learning approach. Finally, the work allows significant improvements in terms of quality of results and computation. For example the approach directly allows the search for sub-optimal solutions or its use within heuristics derived from traditional heuristic methods.
Document type :
Theses
Complete list of metadatas

Cited literature [64 references]  Display  Hide  Download

https://tel.archives-ouvertes.fr/tel-00847745
Contributor : Abes Star <>
Submitted on : Wednesday, July 24, 2013 - 11:56:14 AM
Last modification on : Wednesday, March 27, 2019 - 4:41:29 PM
Long-term archiving on : Friday, October 25, 2013 - 4:10:26 AM

File

VD_RINAUDO_PHILIPPE_05122012.p...
Version validated by the jury (STAR)

Identifiers

  • HAL Id : tel-00847745, version 1

Collections

Citation

Philippe Rinaudo. Algorithmique de l'alignement structure-séquence d'ARN : une approche générale et paramétrée. Autre [cs.OH]. Université Paris Sud - Paris XI, 2012. Français. ⟨NNT : 2012PA112355⟩. ⟨tel-00847745⟩

Share

Metrics

Record views

984

Files downloads

2129