Autonomously Adjusting Multi-Relational Hypergraphs Structure for Predicting circRNA-miRNA Associations

Identifying circRNA-miRNA associations is critical for understanding gene regulatory mechanisms, discovering new biomarkers, and developing therapeutic strategies. The ongoing advancement of autonomous artificial intelligence (AI) technology, particularly in relational and graph learning, enables researchers to develop autonomous AI prediction models to process and analyze existing associations. These models can autonomously extract meaningful patterns and relationships, thereby accurately predicting unknown associations and providing efficient auxiliary tools for traditional experimental methods. Unfortunately, validated reliable circRNA-miRNA associations are often very sparse, making it difficult to learn the intrinsic associations between circRNAs and miRNAs from a static explicit graph structure. To alleviate this problem, we propose a new autonomous AI prediction framework that combines local simple associations with global high-order interactions for joint learning. The framework captures locally embedded representations based on the similarity relationships of RNA molecule attributes, and aggregates contextual information across the global region in different relational modes by applying message passing on multi-relational hypergraphs. Furthermore, we design an autonomous adjustment strategy for multi-relational hypergraphs. This strategy enables adaptive learning of potential node correlations and autonomous construction of hypergraph structures more suitable for downstream tasks, thereby enhancing higher-order relationships between RNA molecules and improving prediction performance and generalization capabilities. Experimental results on three real-world datasets demonstrate that this framework performs excellently in predicting circRNA-miRNA associations, significantly outperforming existing prediction models.