miRNA is an important class of non-coding RNA, which affects a variety of physiological activities by degrading or inhibiting target mRNA. Current computational methods for miRNA–mRNA interaction prediction often overlook extensive information within the full sequence and do not fully capture the relationships between interaction pairs, resulting in suboptimal feature learning. In this paper, we propose a Degree-Aware Graph Fusion Network (DAGFN) for miRNA–mRNA interaction prediction. DAGFN consists of three main components: hybrid feature learning, graph fusion, and miRNA–mRNA pair (MMP) node feature learning. First, conventional descriptors and mathematical descriptors are used to extract hybrid features from full miRNA sequences and mRNA 3 ′ UTR sequences. The topology graph and feature graph are constructed based on hybrid features. Subsequently, they are integrated to obtain a fused graph. Finally, a degree-aware learning mechanism is applied to mitigate degree bias in the fused graph, enabling efficient learning of MMP node features and enhancing miRNA–mRNA interaction prediction performance. In terms of accuracy, DAGFN improves by 8.42% on the D1 dataset compared to the state-of-the-art methods. On the RNA-seq LFC dataset, DAGFN achieves the best performance in predicting high-functional miRNA–mRNA interactions. Further case studies demonstrate the capability of DAGFN to discover potential miRNA–mRNA interactions.
Xie et al. (Sun,) studied this question.