The paper proposes machine-learning pipelines that allow to automatically generating relevant feature spaces for virus detectors, detect the presence of viral modifications in JS-files and scripts in real time, as well as interpret and visualize the machine solution obtained automatically.  It is shown that the best quality metrics will be demonstrated by models of an abstract syntactic tree using binary classifiers based on ensembles of decision tree. The explanation, the solution  automatically generated by the virus detector, is demonstrated.