Fig. 6

Scatter plot representing performance of MF-LOGP and published models on the final test data set (N = 2,713) as a function of the number of features required by the model. (■) Molecular simulations, (●) fragment/topological analysis, (◆) fragment additive + similarity search, (⬟) molecular simulation with neural network, (★) structurally independent atom additive. The bottom left corner represents the region represents high accuracy with the fewest number of features. The top right region represents models with high errors despite having substantial number of features