The interacting mechanism between proteins and peptides is very important.Computational methods can swift-ly pinpoint protein-peptide interaction sites,thereby significantly reducing research and development expenditures.However,current models cannot predict the binding sites with high precision and face challenges in addressing data imbalance issues.To deal with these problems,a novel prediction method named ResPep based on hybrid deep learning methods is proposed in this paper.In terms of dataset processing,K-means clustering algorithm is used to down-sample majority-class samples in the data-set to balance data samples.At the algorithm level,a lightweight hybrid deep learning model is proposed,comprising residual networks,one-dimensional convolutional neural networks,and multi-head attention mechanisms.Furthermore,cost sensitivity learning is integrated in ResPep model.The experimental results show that ResPep achieves better generalization performance compared with most existing methods on the public test dataset TS125.
关键词
蛋白质-多肽/相互作用位点识别/深度学习/不平衡数据/蛋白质序列
Key words
protein-peptide/interaction site recognition/deep learning/imbalanced data/protein sequences
维普
万方数据