OrchAlign: Orchestrated Multimodal Alignment for Gene Expression Prediction.

OrchAlign achieves state-of-the-art gene expression prediction by explicitly modeling fine-grained cross-modal and intra-modal relationships in DNA and epigenomic data.

• Why it matters: Accurate gene expression prediction requires precise integration of diverse biological modalities, but existing methods lack detailed modeling of token-level interactions and structural coherence, limiting their effectiveness.
• What they did: The approach disentangles DNA and epigenomic signals into shared and specific representations, aligns shared features at the token level within regulatory regions, and enforces structural consistency before multimodal fusion using the BiMamba backbone, involving 100% resource availability.
• The result: This framework enhances the fidelity of cross-modal correspondence and intra-modal structure, leading to improved predictive performance and enabling more accurate understanding of gene regulation mechanisms.