Identifying disease-associated circRNAs based on edge-weighted graph attention and heterogeneous graph neural network

Abstract

MotivationCircular RNAs (circRNAs) with varied biological activities are implicated in pathogenic processes, according to new findings. They are regarded as promising biomarkers for the diagnosis and prognosis due to their structural features. Computational approaches, as opposed to traditional experiments, can identify the circRNA-disease connections at a lower cost. Multi-source pathogenesis data can help to reduce data sparsity and infer probable connections at the system level. The majority of available approaches create a homologous network using multi-source data, but they lose the datas heterogeneity. Effective solutions that make use of the peculiarities of multi-source data are urgently needed. ResultsIn this paper, we propose a model (CDHGNN) based on edge-weighted graph attention and heterogeneous graph neural networks for discovering probable circRNA-disease correlations prediction. The circRNA network, miRNA network, disease network and heterogeneous network are constructed based on the introduced multi-source data on circRNAs, miRNAs, and diseases. The features for each type of node in the network are then extracted using a designed edge-weighted graph attention network model. Using the revised node features, we learn meta-path contextual information and use heterogeneous neural networks to assign attention weights to different types of edges. CDHGNN outperforms state-of-the-art algorithms with comparable accuracy, according to the findings of the trial. Edge-weighted graph attention networks and heterogeneous graph networks have both improved performance significantly. Furthermore, case studies suggest that CDHGNN is capable of identifying particular molecular connections and can be used to investigate pathogenic pathways. Contactjxwang@mail.csu.edu.cn