Figure 2

PCA of the compounds annotated as actives against plasmodial DHFR (green) as well as T. gondii DHFR (red). Overall, plasmodial DHFR inhibitors cover a substantial portion of the chemical space occupied by T. gondii DHFR inhibitors. However, some clusters of T. gondii DHFR inhibitors are located in additional chemical space not covered by the plasmodial inhibitors (red boxes). These clusters contain compounds with bicyclic ring systems. By contrast, plasmodial inhibitors only contain unfused rings (green boxes). These observations explain why recall is low (~35%) when plasmodial DHFR inhibitors are excluded from the training set: T. gondii inhibitors do not cover all relevant chemical space, particularly the space occupied by compounds with unfused ring systems.