Fig. 7
From: QSAR-derived affinity fingerprints (part 2): modeling performance for potency prediction
RMSE on the test set for predictive models trained on either rv-QAFFP 440 or rv-QAFFP 1360. The results for the 43 data sets and 50 replicates are shown. Overall, it can be seen that the performance of models trained on rv-QAFFP computed using base models with low and high predictive power (i.e., rv-QAFFP 1360) is comparable to the performance of models trained on rv-QAFFP (i.e., rv-QAFFP 440) computed using only base models showing high predictive power