Gass-metal: um servidor web para identificação de sítios metálicos similares em proteínas baseado em algoritmos genéticos paralelos

Abstract

Metals are present in more than 30% of proteins found in nature and perform important biological functions, in addition they act in the maintenance of protein structure. Metal ions in proteins are bounded to groups of atoms and this set is called a metal-binding site. Metal-binding sites can perform catalytic, structural, transport and electron transfer functions in a protein. Traditional and experimental techniques for metal-binding site prediction usually find obstacles related to time and cost of execution, making computational tools that can assist in predictions become even more important. Several methods in the literature have made efforts to predict metal-binding sites and have shown great results, but they still encounter barriers due to issues related to protein size, type of ions and ligands, ability to find inter-domain residues and even when obtaining not good accuracy rates. The main goal of this master thesis is to adapt GASS algorithm (Genetic Active Site Search), initially proposed for the prediction of catalytic sites, to search for metalbinding sites. The method developed, GASS-Metal, divides residues of a protein in threedimensional space and uses parallelism of genetic algorithms to find candidate sites that are close in relation to the distance of cured templates from M-CSA and MetalPDB. The results of the sanity and homologous protein tests showed that GASS-Metal is a robust method, capable of finding metal-binding sites in different types of ions and does not restrict its search to a single chain. In addition, when using conservative mutations, the prediction accuracy rate improves even more, helping to find sites in situations where it was previously impossible, due to the lack of residues in certain proteins. In comparison to state-of-the-art predictors, GASS-Metal achieved satisfactory performance in predicting metal-binding sites of different ions. The results showed that the method was superior in the prediction in 5 of the 12 metal ions evaluated and still obtained equivalent performance in other 6 different metal-binding sites.