AVALIAÇÃO DE TÉCNICA DE ESPECTROSCOPIA ÓPTICA E APRENDIZADO DE MÁQUINA PARA DIAGNÓSTICO SOROLÓGICO DE Brucella abortus E Mycobacterium bovis EM BOVINOS

Abstract

Tuberculosis and bovine brucellosis are important diseases for cattle farming in Brazil, given the economic impacts due to the sacrifice of positive animals, condemnation of products of animal origin and the risk to public health. In addition, B. abortus and M. bovis are very resistant microorganisms in the environment, which facilitates their dissemination. The National Program for the Control and Eradication of Brucellosis and Tuberculosis (PNCEBT) promotes vaccination against brucellosis, diagnostic tests for B. abortus and M. bovis, sacrifice or sanitary slaughter of positive animals and voluntary adherence actions with the purpose of promoting the prevention and control of these diseases. However, the tests currently employed may have some specificity and/or sensitivity problems, or demand for a complex laboratory structure, high cost and excessive execution time. Thus, research on methods based on spectroscopy such as FTIR, UV-Vis and others is justified, which associated with multivariate analysis methods and machine learning algorithms have shown to be promising in the diagnosis of diseases, presenting indices acceptable sensitivity-specificity, lower costs and faster results. Thus, the UV (Ultraviolet) method associated with machine learning was evaluated in this research for the diagnosis of brucellosis and tuberculosis in cattle. The UV methodology was evaluated for both diseases, with different antigens (B. abortus antigen for serum agglutination test; recombinant antigens P27, MPB83 and MPB70 of M. bovis). A total of 106 bovine serum samples (53 positive and 53 negative) for brucellosis and 88 samples (44 positive and 44 negative) for tuberculosis were used. The antigen-serum ratios were evaluated for each antigen, and it was identified, through principal component analysis (PCA), that the ratio 1:1 for brucellosis and, 1:16 (MPB83), 1:2 (MPB70) and 1:2 (P27), for tuberculosis, showed better separation of positives and negatives. Furthermore, with the proportion defined, the collection of the UV spectra of the total samples was carried out. Then the spectra were submitted to principal component analysis in order to observe the tendency of cluster formation. For brucellosis, no clustering tendency was observed, while for tuberculosis, only the P27 antigen showed good clustering results. Finally, using machine learning algorithms, an overall accuracy of 92.5% for brucellosis and 96.3% for tuberculosis (using P27 antigen) was achieved.