Author(s): Guasco, Luciano M.
Date: 2010
Persistent ID: http://hdl.handle.net/10362/5143
Origin: Repositório Institucional da UNL
Subject(s): Multiple sequence alignment; Molecular coevolution; Mutual information; Constraints; Amino acids correlation
Trabalho apresentado no âmbito do European Master in Computational Logics, como requisito parcial para obtenção do grau de Mestre em Computational Logics
One of the most important fields in bioinformatics has been the study of protein sequence alignments. The study of homologous proteins, related by evolution, shows the conservation of many amino acids because of their functional and structural importance. One particular relationship between the amino acid sites in the same sequence or between different sequences, is protein-coevolution, interest in which has increased as a consequence of mathematical and computational methods used to understand the spatial, functional and evolutionary dependencies between amino acid sites. The principle of coevolution means that some amino acids are related through evolution because mutations in one site can create evolutionary pressures to select compensatory mutations in other sites that are functionally or structurally related. With the actual methods to detect coevolution, specifically mutual information techniques from the information theory field, we show in this work that much of the information between coevolved sites is lost because of mistakes in the multiple sequence alignment of variable regions. Moreover, we show that using these statistical methods to detect coevolved sites in multiple sequence alignments results in a high rate of false positives. Due to the amount of errors in the detection of coevolved site from multiple sequence alignments, we propose in this work a method to improve the detection efficacy of coevolved sites and we implement an algorithm to fix such sites correcting the misalignment produced in those specific locations. The detection part of our work is based on the mutual information between sites that are guessed as having coevolved, due to their high statistical correlation score. With this information we search for possible misalignments on those regions due to the incorrect matching of amino acids during the alignment. The re-alignment part is based on constraint programming techniques, to avoid the combinatorial complexity when one amino acid can be aligned with many others and to avoid inconsistencies in the alignments. In this work, we present a framework to impose constraints over the sequences, and we show how it is possible to compute alignments based on different criteria just by setting constraint between the amino acids. This framework can be applied not only for improving the alignment and detection of coevolved regions, but also to any desired constraints that may be used to express functional or structural relations among the amino acids in multiple sequences. We show also that after we fix these misalignments, using constraints based techniques, the correlation between coevolved sites increases and, in general, the new alignment is closer to the correct alignment than the MSA alignment. Finally, we show possible future research lines with the objective of overcoming some drawbacks detected during this work.
