Multi-View State-Space Modeling for Zero-Shot circRNA-miRNA Interaction Prediction.

MV-S2M, a multi-view state-space model for circRNA–miRNA interaction prediction, achieves state-of-the-art performance in both zero-shot and conventional settings on multiple benchmark datasets.

• Why it matters: Existing CMI predictors largely ignore disease-level regulatory context, limiting their ability to characterize cross-pathway dependencies during zero-shot inference.
• What they did: MV-S2M uses structure-aware diffusion to enrich molecular priors, builds disease-mediated semantic subgraphs, applies state-space modeling for long-range and view-specific patterns, then fuses multi-view representations via linear projection into an MLP.
• The result: Experiments on multiple benchmarks show state-of-the-art results for zero-shot and conventional settings, with cold-start evaluations revealing distinct generalization behavior across unseen-entity settings.