Desenvolvimento de uma ferramenta para identificar proteínas ribossomais em espectro de massa do tipo MALDI-TOF

Abstract

Bacterial identification is a topic of great standard in the field of agriculture for the understanding of soil microbiology, especially the rhizosphere. Among all techniques for identifying microorganisms, Mass Spectrometry MALDI-TOF type has been extensively adopted as a more economical and effective alternative than traditional methods, due to its phenotypic characteristics. This method facilitates microorganism identification, as each microorganism possesses a distinct mass spectrum profile. Within the produced mass spectrum, specific biomarkers can be assigned and utilized as criteria for sample classification. Ribosomal proteins are examples of biomarkers poised for bacterial identification, given their roles in cellular maintenance and their remarkable conservation in amino acid sequences. The Ribopeaks bacterial classifier uses molecular mass data from ribosomal proteins for organism identification. However, mass spectra data obtained from whole bacterial might include peaks associated with non-ribosomal proteins, peptides, metabolites, and lipids within their distinctive patterns, creating challenges for accurate classification. In this study, a clustering approach was employed, utilizing the DBSCAN algorithm, to cluster ribosomal proteins regardless of their specific types. This approach aimed to create a filter capable of determining the compatibility of a given macromolecule mass with a ribosomal protein. For construction of the models, the Puchuy base of presumed masses of ribosomal proteins was used, which went through a pre-processing step before being submitted to machine learning. A controller for multiple bacterial classifications in Ribopeaks was built to enable the validation of the generated models, sending the organisms from the real SpectraBank database to the classifier before and after filtering the peaks. In the best case, the filter was able to subtly increase the assertiveness of the bacterial classifier, with an average reduction of 40.1% in the peak volume of the bacterial sample and a reduction of 35.66% in the processing time for classification of the same organisms.