Infectious bursal disease virus: predicting viral pathotype using machine learning models focused on early changes in total blood cell counts

Abstract

Abstract Infectious bursal disease (IBD) is an avian viral disease caused in chickens by infectious bursal disease virus (IBDV). IBDV strains (Avibirnavirus genus, Birnaviridae family) exhibit different pathotypes, for which no molecular marker is available yet. The different pathotypes, ranging from sub-clinical to inducing immunosuppression and high mortality, are currently determined through a 10-day-long animal experiment designed to compare mortality and clinical score of the uncharacterized strain with references strains. Limits of this protocol lie within standardization and the extensive use of animal experimentation. The aim of this study was to establish a predictive model of viral pathotype based on a minimum number of early parameters measured during infection, allowing faster pathotyping of IBDV strains with improved ethics. We thus measured, at 2 and 4 days post-infection (dpi), the blood concentrations of various immune and coagulation related cells, the uricemia and the infectious viral load in the bursa of Fabricius of chicken infected under standardized conditions with a panel of viruses encompassing the different pathotypes of IBDV. Machine learning algorithms allowed establishing a predictive model of the pathotype based on early changes of the blood cell formula, whose accuracy reached 84.1%. Its accuracy to predict the attenuated and strictly immunosuppressive pathotypes was above 90%. The key parameters for this model were the blood concentrations of B cells, T cells, monocytes, granulocytes, thrombocytes and erythrocytes of infected chickens at 4 dpi. This predictive model could be a second option to traditional IBDV pathotyping that is faster, and more ethical.

Keywords

• Infectious bursal disease
• Gumboro
• pathotype
• predictive model
• machine learning
• blood formula