计算机科学
图形
嵌入
核(代数)
卷积(计算机科学)
源代码
人工智能
计算复杂性理论
算法
机器学习
理论计算机科学
人工神经网络
数学
操作系统
组合数学
作者
Li Wen,Shulin Wang,Hu Guo
标识
DOI:10.1007/978-3-030-91415-8_20
摘要
Predicting lncRNA-protein interactions (LPIs) through computational models can not only help to identify the function of lncRNAs, but also help to solve the problem of huge cost of materials and time. In this study, we develop a novel computational model combining fast kernel learning (FKL) and multi-layer graph convolution network (GCN) to identify potential lncRNA-protein interaction (LPI-FKLGCN). The LPI-FKLGCN model can fuse the multi-source features and similarities by the FKL technique and code the embedding representive vectors by the multi-layer graph convolution network. Through 5-fold cross-validation, the LPI-FKLGCN obtains an AUPR value of 0.52 and an AUC value of 0.96, which is superior to other methods. In case studies, most of the predicted LPIs are confirmed by the newly published biological experiments. It can be seen that the fusion of multi-source similarities and features, combined with multi-layer embedding vectors from graph convolution network can improve the accuracy of LPI prediction and the model of LPI-FKLGCN is an efficient and accurate tool for LPI prediction.
科研通智能强力驱动
Strongly Powered by AbleSci AI