ACRF is more effcient than ACCpro.
With ACRF, the accuracy of SA prediction is improved.
ACRF is robust and easy to use.
Easy to use.
ACRF is developed for predicting solvent accessibility which is based on a high order conditional random field model. This method not only models the interdependency among adjacent residues, but also exploits the correlation among long range residues. First, it is discovered that different secondary structures has different periodicity of solvent accessibility. Therefore, high order features are firstly introduced, and different orders of models are used on different secondary structures. Second, this study achieves improvement by using new features such as sequence conservation, contact numbers. In conclusion, this method has higher accuracy than other existing ones. In addition, through case studies, it is shown that high order terms supplement local features with long range correlations, and correct the wrong positions predicted by bidirectional recurrent neural network and chain condition random field.