Skip to main content
. Author manuscript; available in PMC: 2015 Nov 1.
Published in final edited form as: Proteins. 2014 Sep 25;82(11):3170–3176. doi: 10.1002/prot.24682

Figure 2.

Figure 2

The Pearson’s correlation coefficient for predicting the accessible surface area as a function of the number of neural network instances on which an average is taken. We test again the effect of removal of the global features and find similar results to before. Results were obtained on an independent test set as described in the text. Note that global features allow also for the accuracy to reach a peak after a relatively small number of networks (around 5). In contrast, without global features the networks do not peak even after 24 instances. Note that these correlations are for the ASA values and not normalized ASA.