This work proposes FastMI-HGNet, a two-stream heterogeneous graph neural network for multi-omics disease classification and biomarker prioritization that supports interpretable multi-omics disease classification and biomarker prioritization.
Abstract
Background/Objectives: Multi-omics datasets are increasingly used for disease classification, but differences in scale, distribution, and resolution across omics layers complicate their integration. Conventional fusion approaches may overlook nonlinear cross-omics dependencies and structured sample–feature relationships. Here, we propose FastMI-HGNet, a two-stream heterogeneous graph neural network for multi-omics disease classification. Methods: The framework uses fast mutual information (FastMI) to construct dependency edge priors for a heterogeneous graph that connects sample and feature nodes. A Transformer-based data stream captures vector-level feature interactions, while a graph attention stream models structural dependencies among samples and molecular features. An uncertainty-aware ensemble further improves stability under small-sample and noisy multi-omics settings. Results: Evaluated on five public multi-omics benchmarks—ROSMAP, LGG, BRCA, and the more challenging COAD tumor-stage classification task, together with KIPAN as a ceiling-level proof-of-concept benchmark—FastMI-HGNet achieved competitive classification performance while supporting interpretable biomarker prioritization. In BRCA, SHAP-based analysis highlighted model-prioritized genes such as FOXC1 and SOX10. Conclusions: FastMI-HGNet supports interpretable multi-omics disease classification and biomarker prioritization.
Integrating upstream genomic variations and downstream functional proteomic signalling is necessary for developing precision oncology. There are challenges in the integration of multi-omics data due to large discrepancies in dataset dimensions, structural heterogeneity, and non-linear interdependencies between modalities. All of these challenges can cause high-dimensional data of genomics to overshadow important data of proteomics. To remedy these issues, we developed a new multi-omics integration framework and called it CrossGeneProtein-Net (CGP-Net). CrossGeneProtein-Net utilizes a Sparse Autoencoder (SAE) that compresses the high-dimensional genome data and a Protein-Protein Interaction (PPI) focused Graph Attention Network (GAT) that forms a structure of interrelated biological activities that are incorporated within the proteomic communication framework. Capturing cross talking of the different modalities was accomplished through a bidirectional cross-attention architecture. An InfoNCE objective with a contrastive approach served to keep relative positions of the data according to the degree of intermodal noise and were more easily interpretable and consistent in relation to the data. Validation of CGP-Net was conducted on multiple TCGA and CPTAC datasets and it outperformed baseline models MOGONET and OASIS with statistical significance of p < 0.01. On the TCGA-BRCA benchmark, CGP-Net achieved a macro-balanced accuracy of 91.82% for PAM50 molecular subtyping and an overall survival concordance index (C-index) of 0.784, significantly outperforming state-of-the-art baselines including MOGONET and OASIS (p < 0.01). The Integrated Gradients method of feature attribution validated CGP-Net’s superior performance through its emphasis on important oncogenes as well as pathways.
Sowmya K, Ananth Prabhu G, Mustafa Basthikodi· International journal of com...· 0 citations
Omics data, such as transcriptomics, proteomics, and metabolomics, provide critical insights into disease mechanisms and clinical outcomes. However, their high dimensionality, small sample sizes, and intricate biological networks pose major challenges for reliable prediction and meaningful interpretation. Graph Neural Networks (GNNs) offer a promising way to integrate prior knowledge by encoding feature relationships as graphs. Yet, existing methods typically rely solely on either an externally curated feature graph or a data-driven generated one, which limits their ability to capture complementary information. To address this, we propose the external and generated Graph Neural Network (engGNN), a dual-graph framework that jointly leverages both external known biological networks and data-driven generated graphs. Specifically, engGNN constructs a biologically informed undirected feature graph from established network databases and complements it with a directed feature graph derived from tree-ensemble models. This dual-graph design produces more comprehensive embeddings, thereby improving predictive performance and interpretability. Through extensive simulations and real-world applications to gene expression data, engGNN consistently outperforms state-of-the-art baselines. Beyond classification, engGNN provides interpretable feature importance scores that facilitate biologically meaningful discoveries, such as pathway enrichment analysis. Taken together, these results highlight engGNN as a robust, flexible, and interpretable framework for disease classification and biomarker discovery in high-dimensional omics contexts.
Unknown authors· Briefings in Bioinformatics· 0 citations
Accurate cancer survival prediction is important for risk stratification and personalized treatment. However, patient prognosis is shaped by complex molecular regulation, cross-omics dependencies, and subtype-dependent tumor heterogeneity, making accurate and interpretable prediction challenging. Existing multi-omics survival models often treat omics profiles mainly as feature views and learn coarse patient-level representations, leaving fine-grained gene regulatory topology and subtype-structured prognostic heterogeneity underexploited. We propose LaCONIC, a label-aware and graph-guided multi-omics collaborative learning framework that bridges fine-grained molecular regulation and coarse-grained patient prognosis. LaCONIC first learns topology-aware gene structural representations through heterogeneous regulatory graph pretraining on a multi-entity disease gene regulatory network. It then performs adaptive cross-omics representation learning for intra-omics expression-graph fusion and inter-omics alignment, followed by ceRNA-guided attention to model miRNA-mRNA-lncRNA interactions under biological priors. During training, LaCONIC requires subtype annotations and leverages them through subtype classification and multi-level contrastive constraints to preserve subtype-dependent survival heterogeneity and improve risk-discriminative representation learning. Experiments across multiple TCGA cancer datasets demonstrate that LaCONIC consistently outperforms 14 representative survival baselines, while SHAP-based analyses identify prognosis-associated regulatory modules and biomarkers.
Pei Liu, Xiao Liang, Jia-Wei Luo· Proceedings of the 32nd ACM...· 0 citations
Motivation Graph Neural Networks (GNNs) have gained increasing interest in the biomedical domain, as the integration of prior knowledge and deep neural networks has the potential to enhance insights into molecular processes and disease mechanisms. However, a comprehensive and systematic assessment of model architectures, data modalities, graph structures, and their performance for graph signal classification in the biomedical domain is yet to be performed. In order to close this gap, we conducted a benchmarking study on multiple GNNs on a Protein-Protein Interaction (PPI) network for Kidney Renal Clear Cell Carcinoma and Breast cancer subtype prediction, performing an in-depth investigation of architectures, incorporating skip connections and various data modalities. Results While none of the GNNs outperforms the structure-agnostic Multi-Layer Perceptron baseline, all of them can handle bimodal data (gene methylation and expression) and offer the ability to gain explainability based on PPIs. We offer practical guidelines for applying GNNs to graph signal processing tasks specifically for cancer classification. Depending on the underlying dataset and PPI structure employed, models on different data modalities outperform others. Overall, we suggest using ChebNet, which tends to outperform the Graph Convolutional Network and the Graph Attention Network in cancer subtype prediction. We recommend using GNN architectures that employ a simple flattening readout layer, as they provide better classification performance and faster training time than those with global average pooling. Additionally, we tested residual connections, but they had only an insignificant impact on classification performance. Availability and implementation Code and data available at https://github.com/HauschildLab/GNN4PPI. Contact julia.schirmacher@med.uni-goettingen.de, anne-christin.hauschild@uni-gießen.de Supplementary information Supplementary data attached.
Julia Schirmacher, M. C. Maurer, J. M. Metsch et al.· bioRxiv· 0 citations
Background Weighted Gene Co-expression Network Analysis (WGCNA) is a widely adopted systems biology method to discover gene modules and module-trait associations, mostly from transcriptomics. Designed for a single layer, it cannot jointly analyze multi-omics layers, a consequential limitation in modern biomedical research. WGCNA modules are often hard to interpret, requiring vast follow-up for contextualization. Moreover, no integrated framework exists to visualize condition-specific, cross-omics relationships at module or feature level. Results To address these limitations, we developed WGCNA+, a novel R package extending WGCNA to multi-omics. WGCNA+ offers key innovations: (i) a unified multi-omics pipeline for per-layer network inference and cross-layer module enrichment; (ii) SVD-accelerated topological overlap matrix calculation that greatly reduces computation time; (iii) a consensus framework identifying modules reproducible across independent datasets/conditions; (iv) LASAGNA, a companion R package for phenotype-conditioned, multi-partite graph visualization of cross-omics relationships; (v) AI-powered annotation and infographics offering immediate biological insight. We tested WGCNA+ across public transcriptomics, proteomics, and miRNA datasets. WGCNA+ detects biologically meaningful modules, cross-omics feature and phenotype correlations, and provides AI-powered interpretation that accelerates research. Conclusions WGCNA+ addresses existing gaps with a principled, efficient framework for co-expression network analysis across omics. It detects cross-omics regulatory modules and their phenotype association to support basic research, biomarker discovery and pathway analysis. It uniquely offers AI-assisted interpretation and infographics, aiding hypothesis generation. Complementing WGCNA+, LASAGNA is a phenotype-aware multi-partite visualization framework to explore cross-omics relationships. Altogether, these features make WGCNA+ an innovative, powerful tool for clinical and translational research. Availability and implementation WGCNA+ and LASAGNA are implemented in R language for statistical computing, version≥ 3.5. WGCNA+ and LASAGNA are fully and freely available with no restrictions (https://github.com/bigomics/WGCNAplus; https://github.com/bigomics/lasagna).
Antonino Zito, Xavier Montagut, Santiago Cano-Muniz et al.· bioRxiv· 0 citations
Pan-cancer multi-omics analysis requires models that integrate complementary molecular signals while preserving biologically meaningful relationships among patients. This study presents a leakage-controlled benchmarking framework for patient-graph learning in pan-cancer classification and prognosis analysis, focusing on how graph construction affects downstream performance. The benchmark explicitly separates fold-specific graph formation from downstream prediction. Using the TCGA Pan-Cancer cohort of 8204 primary tumors across 31 cancer types, RNA expression and copy-number variation data were used to compare early feature fusion, lightweight similarity network fusion (SNF-lite), and fused k-nearest neighbor similarity graphs under a common GATv2 encoder family with a matched attention-head search space and inner-validation selection procedure. A strict 5 × 3 nested cross-validation protocol ensured that imputation, gene selection, feature scaling, similarity computation, and neighbor search were fitted on training folds only. At G'=2000, graph-level fusion approaches achieved about 0.92 accuracy and 0.89 Macro-F1, outperforming early fusion at about 0.89 accuracy and 0.84 Macro-F1. Fused kNN graphs also showed higher neighborhood label purity than SNF-lite despite similar predictive performance. A weighted topology audit showed that local label agreement alone did not determine graph utility. Gene and omics ablations showed that RNA carried the dominant subtype-discriminative signal, while CNV and mutation contributed weaker but complementary information. A Cox auxiliary objective retained classification performance when used alone and enabled out-of-fold prognostic stratification. These findings show that patient-graph construction is a key design choice in pan-cancer multi-omics learning and that leakage-controlled evaluation is essential for reliable and biologically informative benchmarking in computational oncology.
