Protein Secondary Structure Prediction Using Deep Convolutional Neural Fields

Abstract

Protein secondary structure (SS) prediction is important for studying protein structure and function. When only the sequence (profile) information is used as input feature, currently the best predictors can obtain ~80% Q3 accuracy, which has not been improved in the past decade. Here we present DeepCNF (Deep Convolutional Neural Fields) for protein SS prediction. DeepCNF is a Deep Learning extension of Conditional Neural Fields (CNF), which is an integration of Conditional Random Fields (CRF) and shallow neural networks. DeepCNF can model not only complex sequence-structure relationship by a deep hierarchical architecture, but also interdependency between adjacent SS labels, so it is much more powerful than CNF. Experimental results show that DeepCNF can obtain ~84% Q3 accuracy, ~85% SOV score, and ~72% Q8 accuracy, respectively, on the CASP and CAMEO test proteins, greatly outperforming currently popular predictors. As a general framework, DeepCNF can be used to predict other protein structure properties such as contact number, disorder regions, and solvent accessibility.

Protein Secondary Structure Prediction Based on Deep Learning

“…Moreover, β is a scalar. In [5], the convolutional layers are configurable to their filter size k and number of filters d, which determine the structure of the CNN and affect the training performance of the model to a certain extent.…”

“…Performance comparison with other network models. Zhou & Troyanska,2014[3] 66.4 SØnderby et al2014 [4] 67.4 Deep-CNF.2016 [5] 68.3 This paper 69. 4 …”

“…In 2014, Zhou and Troyanska [3] proposed the use of a convolution of the Generative Stochastic Network (GSN) method, and then Søren Kaae Sønderby et al [4] proposed a method based on Bi-directional LSTM, which obtained 66.4%, 67.4% Q8 accuracy respectively. In 2016, a method of the deep convolutional neural field (Deep-CNF) obtained 68.3% Q8 accuracy in [5].…”

Redes neurais residuais profundas e autômatos celulares como modelos para predição que fornecem informação sobre a formação de estruturas secundárias proteicas

“…No entanto, futuramente poderemos adaptar um modelo similar que utilize PSSM. Nesse ponto, é relevante mencionar que todos os 5 mé-todos testados por Wang e colaboradores [71] e que utilizam PSSM, entre eles o PSIPRED e a rede neural convolucional profunda desenvolvida por eles, tem uma perda de até 10% de acurácia quando o alinhamento para construção da PSSM contém poucas proteínas. Para esses casos, os 5 métodos apresentaram acurácia em torno de 74% [71].…”

“…Para esses casos, os 5 métodos apresentaram acurácia em torno de 74% [71]. [71]. Para os demais casos, a acurácia foi calculada utilizando nossos dados e com os resíduos que apresentam consenso na atribuição da estrutura secundária.…”