基于稠密网络的DNA结合蛋白质预测方法

Prediction of DNA-Binding Protein Based on Dense Network

提出了一种基于深度学习的DNA结合蛋白识别方法(DBP-DenseNet)。以稠密网络代替传统的金字塔式卷积神经网络(CNN)结构,将上一层的特征信息整合到下一层,利用层间的特征融合来学习整个氨基酸序列的综合特征;由双向长短期记忆网络(Bi-LSTM)负责在氨基酸序列上下文中获取长期依赖关系。实验结果表明,稠密连接可以有效提高模型的特征表达能力。

In this paper,a DNA binding protein recognition method based on deep learning(DBP DenseNet)is proposed.A dense network is used to replace the traditional pyramid convolutional neural network(CNN)structure.The feature information of the upper layer is integrated into the next layer,and the comprehensive features of the whole amino acid sequence are learned by the feature fusion between layers.The Bi-LSTM is responsible for obtaining the long-term dependence in the context of amino acid sequences.Experimental results show that dense connection can effectively improve the ability of feature representation.