Identificação de proteínas ribossomais em espectro de massa do tipo Maldi-Tof

Abstract

Ribosomal proteins are used as biomarkers in the taxonomic identification of bacteria found in the 50S and 30S subunits based on their protein expression in the MALDI-TOF mass spectrum. It is possible to determine the bacterial taxon, with the clash between known and unknown spectra. Having applications in several areas, such as agronomy in which they require the identification of new microorganisms that assist in plant growth. Thus the identification of ribosomal proteins from the statistical distributions was necessary. To obtain a probabilistic recognition model of possible ribosomal proteins. Ahead it was necessary to build a Generalized Linear Model - GLM, accompanied by the Genetic Algorithm, which stands out for its simplicity, robustness in solving complex binary problems and in the selection of variables for training. For the distinction of the best classification models, the lowest information criterion of Akaike - AIC was taken into account. The results were 59 histograms of ribosomal proteins. Thus eight statistical distributions were selected for the test and one Laplacian mixture model, some distributions were promising when analysed by the statistical tests and yet some data showed with several peaks. The genetic algorithm was favourable in the search of protein combinations for the training of the classificatory model. These combinations generated two models, the first with better accuracy and the other more specific with lower accuracy. The methodology elaborated in this study presented as an alternative to the discrimination of ribosomal proteins from the GLM classifier.