The comparison of performances of prediction models including drug targets' GO term annotation. a) This broken line graph illustrates the performances when the prediction models used different levels of GO term annotations. A red line represents results using SVM, while a blue means represents results using logistic regression. The x-axis shows the different threshold levels of the GO terms. "No Net" means the prediction models used ADR vs. drug target facts; "LvN" means we just used a GO term equal to or deeper than level N in the GO hierarchy (we defined the term of "biological process" as level 1). b) The x-axis shows each level of the GO terms. "LvN" means we replaced the GO terms deeper than level N with their level N ancestors.