基于神经网络的LncRNA与蛋白质互作关系预测算法

An Algorithm for Predicting LncRNA-protein Interactions Based on Neural Networks

论文提出一种基于注意力机制(Attention)的融合神经网络预测方法预测LncRNA与蛋白质的相互作用,命名为PIPAFNN。通过栈式自编码器和融合神经网络(CNN)-长短期记忆网络(LSTM)分别对LncRNA和蛋白质的序列进行特征提取,在模型学习过程中使用注意向量,使得训练出的模型能够关注不同样本中对预测方法具有更大影响的特征属性,从而有效地预测LncRNA和蛋白质的互作关系。同时,利用五折交叉验证,模型在拟南芥和玉米数据集上的AUC值分别是0.9582和0.9251,与其他机器学习方法进行比对提升7%和2%,模型的分类效果更为显著。

In this paper,an attention-based fusion neural network prediction method is proposed for predicting LncRNA-Protein interactions,named PIPAFNN.The model uses Stacked Autoencoder and fusion neural network(CNN)-long short-term memory(LSTM)) to extract features from LncRNA and protein sequences respectively,and uses attention vectors in the model learning process,so that the trained model can focus on feature attributes in different samples that have a greater impact on the prediction method,thus effectively predicting LncRNA-Protein interactions.Moreover,by using 5-fold cross-validation,the AUC values of the model on the Arabidopsis thaliana and Zea mays datasets are 0.9582 and 0.9251 respectively,which are improved by 7% and2% when compared with other machine learning methods,and the model is more effective in classification.