Multimodal learning is increasingly pivotal in biomedical research, where multi-omics technologies enable comprehensive characterization of diverse molecular layers. Despite their potential, effective integration of heterogeneous multi-omics data remains challenging due to high dimensionality, modality inconsistency, and complex inter-sample dependencies. Existing methods primarily focus on intra-sample feature fusion, often overlooking high-order structural relationships across samples, which limits their ability to capture system-level interactions. To address these challenges, we propose HyperGraph-Enhanced MultiModal Dynamics (HGEMMD), a novel framework that simultaneously models intra-sample modality fusion and inter-sample high-order associations for robust clinical multi-omics integration and classification. HGEMMD incorporates a multimodal dynamics module to alleviate data sparsity and modality heterogeneity, and constructs a modality-aware hypergraph where each hyperedge connects semantically or functionally related samples, capturing non-pairwise dependencies. To further enhance robustness and structural coherence, we introduce relational consistency learning that aligns sample-level relational patterns before and after hypergraph propagation, preserving local semantics while ensuring global structural information. Extensive experiments on benchmark multi-omics datasets demonstrate that HGEMMD consistently outperforms state-of-the-art approaches, validating its effectiveness in robust and trustworthy clinical multi-omics integration. Our code and data are available at https://github.com/ljw-struggle/HGEMMD.

RiboNucleic Acid (RNA) editing is a dynamic and essential biological process that has multifaceted functions in gene regulation, protein diversity, and immune response. Tissue-specific RNA editing is governed by the presence and activity of RNA editing enzymes, such as adenosine deaminase acting on RNA enzymes, and is influenced by the cellular context and regulatory factors in each tissue. As a result, RNA editing can exhibit tissue-specificity. To fully understand the functional implications of RNA editing, it is important to consider its tissue-specific nature and its potential impact on the biology of specific tissues and organs. Utilizing convolutional neural networks, we designed models that can predict RNA editability. The models were validated independently using Clustered Regularly Interspaced Short Palindromic Repeats (CRISPR)/CRISPR associated protein 9 (Cas9)-based Adenosine Deaminase Acting on RNA (ADAR) knockout in both Jurkat and Human Embryonic Kidney 293T‌ (HEK293T) cells. Although RNA editing can be categorized into Alu and non-Alu RNA editing, with the majority of RNA editing falling within the Alu category, our motif and phylogenetic analyses reveal that the tissue-specific characteristics of RNA editing are primarily attributed to non-Alu-related RNA editing. Based on these results, we developed a web server that incorporates RNA editability prediction models for 30 distinct tissue types in Humans and four other species (mouse, bee, fly, and squid). This tool assists studies that aim to gain a more comprehensive understanding of RNA editing-related gene regulation, cellular diversity, and the molecular basis of tissue-specific diseases.

Prediction of enhancer-promoter interactions (EPIs) is key to regulating gene expression and diagnosing genetic diseases. Due to limited resolution, biological experiments perform not as well as expected while precisely identifying specific interactions, giving rise to computational biology approaches. Many EPI predictors have been developed, but their prediction accuracy still needs to be enhanced. Here, we design a new model named EPIMR to identify enhancer-promoter interactions. First, Hilbert Curve is utilized to represent sequences to images to preserve the position and spatial information. Second, a multi-scale residual neural network (ResNet) is used to learn the distinguishing features of different abstraction levels. Finally, matching heuristics are adopted to concatenate the learned features of enhancers and promoters, which pays attention to their potential interaction information. Experimental results on six cell lines indicate that EPIMR performs better than existing methods, with higher area under the precision-recall curve (AUPR) and area under the receiver operating characteristic (AUROC) results on benchmark and under-sampling datasets. Furthermore, our model is pre-trained on all cell lines, which improves not only the transferability of cross-cell line prediction, but also cell line-specific prediction ability. In conclusion, our method serves as a valuable technical tool for predicting enhancer-promoter interactions, contributing to the understanding of gene transcription mechanisms. Our code and results are available at https://github.com/guofei-tju/EPIMR.