Skip to main content
. 2013 Feb 21;9(2):e1002899. doi: 10.1371/journal.pcbi.1002899

Figure 2. Model validation.

Figure 2

(A,B,C) Our models perform robustly in both internal validation (unknown combinations of known drugs and known mutants) and (D,E,F) external validation (unknown combinations of drugs and mutants, one of which is unknown). The PIs perform the best (RMSE 0.27 log units, CCP 93% internal and 0.43 log units, CCP 90% external), followed by the NNRTIs (RMSE 0.45 log units, and CCP 93% internal and 0.49 log units, CCP 91% external) and then the NRTIs (RMSE 0.31 log units, CCP 80% internal and 0.52 log units, CCP 68% external). The range of Log FC values present in the dataset is the largest for the NNRTIs, followed by the PIs and then the NRTIs.