An Explainable and Scalable Hierarchical Tensor Framework for Integrating Multi-Omics and Digital Pathomics
Authors: Ahmad Hussein, Ali Anaissi, Mukesh Prasad, Ali Braytee
Organizations: School of Computer Science, University of Technology Sydney, Sydney, NSW, Australia. · TD School, University of Technology Sydney, Sydney, NSW, Australia. · School of Computer Science, The University of Sydney, Sydney, NSW, Australia.
Background: Precision oncology requires integration of heterogeneous molecular and pathology data, yet many fusion models offer limited biologically grounded interpretability and struggle with missing modalities. Methods: We present Hierarchical Interpretable Tensor Fusion (HIT-Fuse), integrating six modalities: somatic copy number variation, DNA methylation, RNA sequencing, microRNA expression, RPPA protein abundance, and vision transformer-derived pathomics across four TCGA cohorts (glioma, lung adenocarcinoma, breast, and kidney cancer). Modality-specific autoencoders compress high-dimensional inputs, followed by two-stage CP tensor decomposition. Missing latent modalities are reconstructed using clinical covariates and observed omics to expand cohorts and retrain downstream models. A mathematical traceback mechanism enables global and local xAI attributions. Results: HIT-Fuse identifies distinct latent subgroups with significant survival stratification in glioma, breast, and kidney cohorts, both with complete and reconstructed modalities. Cohort expansion yields cohort-specific gains, including a 56.4% relative increase in weighted F1 for glioma. Biological validation against DepMap CRISPR screens, miRTarBase and RPPA annotations supports decoded drivers across all four cohorts. Essential gene rates range from 6.2% in glioma to 7.6% in lung cancer; example drivers include ANAPC11, SNRPD1, PLK1 and RPL14. Of decoded miRNAs, 71-92% have miRTarBase regulatory support, while 77-84% of selected genes are covered in DepMap. Conclusions: HIT-Fuse combines hierarchical tensor fusion, missing-modality reconstruction and traceable attributions to support interpretable multi-omics and digital pathology-based cancer prognosis across heterogeneous cohorts.
Figures & tables
Figure 1: Overview of our proposed Hierarchical Multimodal Tensor Fusion framework (HIT-Fuse). The framework consists of four phases: modality-specific encoding and missing-modality latent reconstruction; biologically structured hierarchical tensor factorization; predictive evaluation using Fglobal for subtype classification and clustering-based survival analysis; and structured explainability through global biological validation and local patient-level interpretation.
Table 2
Figure 2: Predictive and prognostic validation of the HIT-Fuse hierarchical tensor framework. (A) Best subtype classification performance (Weighted F1) per cancer under E1 and E2, with the dominant classifier indicated in parentheses. (B) Survival discrimination measured by C-index from the Cox survival head. (C) Effect of cohort completion (E2 − E1) on subtype and survival metrics, illustrating cancer-specific stability patterns.
Figure 3: Global explainability under the standard setting. Left: Fractional contribution of biological layers (DNA, RNA, protein–imaging) from absolute meta-level loadings B(HIT−Fuse) aggregated over meta-factors and normalized per cohort. Right: Frequency of top-ranked meta-factors across group–modality combinations, pooled across cohorts (top-10 per cohort ranked by permutation importance).
E1
E2
Cancer
Accuracy
F1(w)
C
Accuracy
F1(w)
C
LR
RF
SVM
LR
RF
SVM
LR
RF
SVM
LR
RF
SVM
BRCA
0.417
0.701
0.677
0.474
0.644
0.596
0.643
0.421
0.764
0.767
0.502
0.710
0.708
0.605
GLIOMA
0.590
0.584
0.597
0.575
0.567
0.519
0.840
0.508
0.716
0.699
0.561
0.674
0.652
0.752
KIDNEY
0.941
0.934
0.941
0.941
0.934
0.941
0.619
0.914
0.939
0.922
0.918
0.940
0.924
0.722
LUAD
0.319
0.646
0.635
0.387
0.574
0.494
0.600
0.183
0.595
0.636
0.236
0.495
0.495
0.598
Table 3: Detailed HIT-Fuse hierarchical tensor framework performance across subtype classifiers and survival discrimination. The best F1(w) per cohort is highlighted in bold. C-index corresponds to the survival head.
Figure 4: Latent-space clustering and survival stratification under HIT-Fuse (E2, K=3 ). For each cohort, patients are clustered in the learned latent space (left within each panel) and evaluated via Kaplan–Meier survival separation (right). Survival is truncated at 3000 days for comparability across cohorts.
Table 4: Representative highest-importance meta-factor per cancer cohort under the standard setting. Only the top five decoded drivers are shown for brevity.
Figure 5: Biological interpretation and external validation of HIT-Fuse-derived drivers. Top A: Cross-cancer enrichment fingerprint across GO Biological Process, KEGG, Reactome, miRTarBase, and miEAA libraries. Heatmap intensity reflects −log10(adjustedp) , and the upper bar indicates the number of cancers in which each term is significantly enriched. Bottom B: External functional validation using DepMap CRISPR essentiality screens and regulatory evidence. Bars indicate overall DepMap gene coverage, essential gene proportion, RPPA-supported genes, and miRTarBase-supported miRNAs.
Cancer
Cohort
Concat_All
M1_GroupCP
M2_FlatCP
HIT-Fuse
Breast
E1
0.736
0.666
0.617
0.644
Breast
E2
0.775
0.716
0.657
0.710
Glioma
E1
0.551
0.574
0.583
0.575
Glioma
E2
0.726
0.698
0.688
0.674
Kidney
E1
0.973
0.958
0.921
0.941
Kidney
E2
0.960
0.941
0.924
0.940
Table 5: Maximum weighted F1 scores achieved under each fusion strategy across cancers and cohorts. For each fusion method, the best-performing subtype classifier (among evaluated models) is reported. Bold indicates the highest F1 within each cancer–cohort setting.
We introduce ProtoPathway, an interpretable-by-design multimodal framework for cancer survival prediction that unifies whole slide imaging and transcriptomics through encoders producing biologically grounded representations on both sides of the fusion. On the histopathology side, K learnable morphological prototypes, trained end-to-end with the survival objective, serve as the slide representation itself: patches flow into prototype tokens via soft assignment, compressing variable-length patch sets into fixed task-adaptive tokens. On the genomic side, a bipartite graph neural network encodes gene expression within the Reactome pathway hierarchy, producing pathway embeddings that reflect both constituent genes and their broader biological context through bidirectional message passing over a shared gene--pathway graph. Cross-modal attention then operates over a compact prototype × pathway matrix in which prototypes query pathways, modeling the biological direction in which molecular programs give rise to tissue morphology. Because both axes carry stable task-learned identity, the attention matrix is itself an interpretability output, yielding native inference-time attribution across the full biological hierarchy, from genes through pathways and prototypes to spatial tissue maps. We evaluate on five TCGA cancer cohorts, demonstrating competitive or superior survival prediction with substantially improved biological interpretability and reduced computational cost, with interpretability claims validated through fold-stratified rank-based population-level analysis. Our source code, model weights, and Reactome pathways, together with a unified codebase reimplementing all multimodal survival baselines under identical preprocessing and evaluation, are available at: https://github.com/AmayaGS/ProtoPathway.
Amaya Gallagher-Syed, Costantino Pitzalis, Myles J. Lewis +2
Cancer survival prediction from multi-omics data remains challenging because prognostic signals are high-dimensional, heterogeneous, and distributed across interacting genes and pathways. We propose PathMoG, a pathway-centric modular graph neural network for multi-omics survival prediction. PathMoG reorganizes genome-scale inputs into 354 KEGG-informed pathway modules, introduces a Hierarchical Omics Modulation module to condition gene-expression representations on mutation, copy number variation, pathway, and clinical context, and uses dual-level attention to capture both intra-pathway driver signals and inter-pathway clinical relevance. We evaluated PathMoG on 5,650 patients across 10 TCGA cancer types and observed consistent improvements over representative survival baselines. The framework further provides gene-level, pathway-level, and patient-level interpretability, supporting biologically grounded and clinically relevant risk stratification.
Di Wang, Chupei Tang, Junxiao Kong +3
1Cuiying Honors College, Lanzhou University, China · School of Mathematics and Statistics, Lanzhou University, China · School of Mathematics and Statistics, Lanzhou University, China and Shanghai Innovation Institute +1
The increase in high-dimensional multiomics data demands advanced integration models to capture the complexity of human diseases. Graph-based deep learning integration models, despite their promise, struggle with small patient cohorts and high-dimensional features, often applying independent feature selection without modeling relationships among omics. Furthermore, conventional graph-based omics models focus on homogeneous graphs, lacking multiple types of nodes and edges to capture diverse structures. We introduce a Heterogeneous Graph ATtention network for omics integration (HeteroGATomics) to improve cancer diagnosis. HeteroGATomics performs multimodal feature selection through a multi-agent system, creating dedicated networks of feature and patient similarity for each omics modality. These networks are then combined into one heterogeneous graph for learning holistic omics-specific representations and integrating predictions across modalities. Experiments on three cancer multiomics datasets show that HeteroGATomics outperforms baseline methods across most evaluation metrics in cancer diagnosis. External validation on an independent breast cancer cohort further demonstrates its generalizability, where HeteroGATomics outperforms graph-based multiomics integration methods across evaluation metrics. Moreover, HeteroGATomics enhances interpretability by identifying important biomarkers contributing to the diagnosis outcomes.
Sina Tabakhi, Charlotte Vandermeulen, Ian Sudbery +1
School of Computer Science, The University of Sheffield, Sheffield, UK · Centre for Machine Intelligence, The University of Sheffield, Sheffield, UK · School of Biosciences, The University of Sheffield, Sheffield, UK