Protein secondary structure (SS) prediction is important for studying proteinstructure and function. When only the sequence (profile) information is used asinput feature, currently the best predictors can obtain ~80% Q3 accuracy, whichhas not been improved in the past decade. Here we present DeepCNF (DeepConvolutional Neural Fields) for protein SS prediction. DeepCNF is a DeepLearning extension of Conditional Neural Fields (CNF), which is an integrationof Conditional Random Fields (CRF) and shallow neural networks. DeepCNF canmodel not only complex sequence-structure relationship by a deep hierarchicalarchitecture, but also interdependency between adjacent SS labels, so it ismuch more powerful than CNF. Experimental results show that DeepCNF can obtain~84% Q3 accuracy, ~85% SOV score, and ~72% Q8 accuracy, respectively, on theCASP and CAMEO test proteins, greatly outperforming currently popularpredictors. As a general framework, DeepCNF can be used to predict otherprotein structure properties such as contact number, disorder regions, andsolvent accessibility.
展开▼
