Multi-Hop Diffusion-Wavelet Semantic Learning for Zero-Shot Circrna-Mirna Interaction Prediction

Abstract

Circular RNAs (circRNAs) and microRNAs (miRNAs) are key regulators in various biological processes, and identifying their interactions is vital for elucidating disease mechanisms. However, due to the high experimental cost and limited coverage of known circRNA-miRNA interactions (CMIs), there is a pressing need for computational approaches that can predict novel interactions beyond the scope of existing data. In this paper, we propose BioWave, a multi-hop diffusion-wavelet model that learns both structural and semantic representations for accurate CMI prediction. Specifically, BioWave employs Transformer-based encoders to extract deep biological semantic features from circRNA and miRNA sequences. In parallel, a multi-hop diffusion-wavelet modeling mechanism grounded in spectral graph theory captures disease-mediated topological dependencies across the interaction network. The semantic and structural embeddings are concatenated and fed into a multi-layer perceptron (MLP) for CMI prediction. Experimental results on three datasets demonstrate that BioWave achieves superior performance compared to state-of-the-art baselines, particularly in zero-shot settings. Case studies further validate its biological interpretability and generalization capability.