One of the principal functions of circular RNA (circRNA) is to participate in gene regulation by sponging microRNAs (miRNAs). Using accumulated circRNA-miRNA associations (CMAs) to construct computational models for predicting potential associations provides a crucial tool for accelerating the validation of reliable associations through traditional experiments. Nevertheless, the current prediction models are constrained in their capacity to represent the higher-order relationships of CMAs and thus require further enhancement in terms of their predictive efficacy. In order to address this issue, we propose a new model based on multirelational hypergraph representation learning (MRHRL). This model employs hypergraphs to capture various higher-order relationships among RNAs and aggregates complementary information through a view attention mechanism. Furthermore, MRHRL introduces a hyperedge-level reconstruction task, jointly optimizing the prediction and reconstruction tasks within a unified framework to uncover potential information, thereby enhancing the model’s predictive and generalization capabilities. Experiments conducted on three real-world data sets demonstrate that MRHRL achieves satisfactory results in CMAs prediction, significantly outperforming existing prediction models.

中文翻译：

---

用于预测 circRNA-miRNA 关联的多关系超图表示学习


环状 RNA （circRNA） 的主要功能之一是通过海绵 microRNA （miRNA） 参与基因调控。使用累积的 circRNA-miRNA 关联 （CMA） 构建用于预测潜在关联的计算模型，为通过传统实验加速验证可靠关联提供了重要工具。然而，目前的预测模型在表示 CMA 的高阶关系的能力方面受到限制，因此需要进一步增强其预测效能。为了解决这个问题，我们提出了一种基于多关系超图表示学习 （MRHRL） 的新模型。该模型采用超图来捕获 RNA 之间的各种高阶关系，并通过视图注意力机制聚合互补信息。此外，MRHRL 引入了超边缘级别的重建任务，在统一的框架内共同优化预测和重建任务，以发现潜在信息，从而增强模型的预测和泛化能力。在三个真实数据集上进行的实验表明，MRHRL 在 CMA 预测方面取得了令人满意的结果，明显优于现有的预测模型。