Prediction of protein interactions in rice and blast fungus using machine learning

Magnaporthe grisea (M.grisea) is the most destructive pathogen cause loss of rice crops every year worldwide. Protein-protein interaction (PPIs) of rice and M.grisea is the key factor of this disease. In this study, an efficient machine learning model is developed to predict the interaction between rice and blast fungus on a genome-scale. From interolog and domain-based method, we obtained 58,379 PPIs. The predicted PPIs are used to developed machine learning model. Testing accuracy of 5-fold cross-validation for these potential PPIs is 88 % and 89 % respectively using amino acid composition (AAC) and conjoint triode (CTD) features. Further the models are tested with other host-pathogen datasets. It predicted fewer PPIs for other host-pathogen databases which confermed that model is specific to rice and blast fungus. The current research work may be a useful resource to plant community to characterize the host-pathogen interaction between rice and M.grisea.