Validação de um método para predição de redes de interação proteína-proteína e sua aplicação em Corynebacterium pseudotuberculosis para identificar proteínas essenciais

Abstract

Corynebacterium pseudotuberculosis (cp) belongs to the group CMNR (Corynebacterium, Mycobacterium, Nocardia, Rhodococcus), is a gram-positive facultative intracellular pathogenic bacterium, have fimbriae, is non-motile, do not form capsules and not sporulate, is presented in serovar ovis and equi. The serovar equi infects horses and cattle. The serovar ovis mainly infects herds of sheep and goats, and is the etiological agent of caseous lymphadenitis (CLA). Cp is prevalent in many countries, causing significant economic losses due to poor quality carcasses decrease in the production of meat, wool and milk. Methods for diagnosis and treatment of CLA are not yet effective enough due Cp have low therapeutic response and ability to persist in the environment, making it an important organism to be researched and understood the systemic level. In this regard, knowing the proteins and their interactions is crucial to understand the molecular mechanisms of the cell, being protein-protein interaction networks an important tool for this type of study. Aiming to generate the Cp interaction network, we worry about validate a methodology for the prediction of interactions with experimental and cured data publicly available. As a result, in addition to increasing the coverage of the network, we obtained an area under the curve (AUC) between 0.93 and 0.96, representing the cutoff of 0.70 a specificity of 0.95 and a sensitivity 0.90. With the validated methodology, the interaction networks were generated for nine serovar ovis Cp strains, being ~99% of interactions mapped from Corynebacterium gender, possessing 15,495 interactions conserved between strains. The shortest path and the degree interaction distribution analysis suggests the predicted networks have biological characteristics. Additionally, we compared the values of the clustering coefficient, Correlation and R2 against randomly generated networks and submit the networks generated to the Shapiro-Wilk normality test. All results show that the predicted interaction networks do not have a random distribution, suggesting the networks were not formed by spurious interactions, existing biological bias its prediction. With validated network, we selected the first 15% of the proteins with more interactions and we identified 181 essential proteins. Only the protein DNA repair protein (RecN) had no homology against database of essential genes (DEG) and other three had homology in just one DEG organism: Catalase (KatA), Endonuclease III (Nth) and trigger factor (Tig ), suggesting they may be good targets for diagnosis and drug development.