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 1.

Figure 1

The mean absolute error in percent for predicting the normalized accessible surface area as a function of the input window size used to train the neural network. The best accuracy is achieved with all inputs at a window size of 21. In addition we remove each input and test the accuracy of the resulting server on the same independent test set. We see that removal of global features reduces the accuracy the most, indicating their significance.