Deep learning in tumour genomics: from multi-omics integration to precision oncology

Clinical Snapshot

Key Findings

• P Value: Not applicable at review level — individual study p-values referenced narratively but not synthesised

• Effect Size: No pooled effect size reported — individual study performance metrics (e.g., AUC, accuracy, F1 scores) are referenced narratively but not aggregated. The review asserts DL 'significantly enhances' detection sensitivity and subtype stratification without providing quantified summary estimates

• Primary Outcome: Qualitative synthesis of DL performance across four tumour genomics domains: (1) DNA sequencing for mutation detection with enhanced sensitivity; (2) gene expression profiling for cancer subtype classification with improved stratification; (3) methylation analysis for epigenetic characterisation; (4) multi-omics integration for comprehensive tumour profiling and therapeutic target identification

• Nnt Or Sensitivity: No NNT, pooled sensitivity, or pooled specificity calculated. Individual DL model performance metrics are discussed qualitatively across cancer types and architectures without aggregation. No diagnostic accuracy meta-analysis (e.g., SROC curves) is presented

• Confidence Interval: Not applicable — no meta-analytic pooling performed; no confidence intervals for pooled estimates are provided

Clinical Application

Clinical implementation of DL-based genomic tools faces substantial feasibility barriers: (1) computational infrastructure requirements exceed most routine clinical laboratory settings; (2) model interpretability remains insufficient for regulatory approval and clinician trust; (3) prospective clinical validation in diverse patient cohorts is largely absent; (4) data governance, patient consent frameworks, and interoperability standards for federated learning are underdeveloped in most health systems. Near-term feasibility is highest in large academic cancer centres with existing bioinformatics infrastructure In Australia, comprehensive tumour genomic profiling is available through the Australian Genomics Health Alliance and select state-based programs (e.g., Peter MacCallum Cancer Centre, Garvan Institute). The TGA has not yet approved any DL-specific genomic diagnostic tool as a standalone clinical decision support system. The MBS does not currently fund DL-augmented genomic analysis as a distinct item. RACGP and COSA (Clinical Oncology Society of Australia) guidelines emphasise evidence-based genomic testing but do not yet reference DL-specific methodologies. The Australian Government's National Cancer Plan 2023–2028 identifies precision medicine as a priority, creating a policy environment receptive to future DL integration pending clinical validation. Data sovereignty considerations under the Australian Privacy Act and the My Health Record framework would need to be addressed before federated learning approaches could be implemented across Australian health networks. Indigenous genomic data representation in training datasets — a known gap in global genomic databases — is a specific equity concern for Australian implementation. Primarily applicable to cancer researchers, bioinformaticians, and computational oncologists rather than direct clinical application. Indirectly relevant to oncologists and pathologists working with tumour genomic profiling platforms in academic cancer centres. Most relevant for patients with cancers where comprehensive genomic profiling is already standard of care (e.g., non-small cell lung cancer, colorectal cancer, breast cancer, haematological malignancies)

Abstract

Cancer remains a leading cause of death globally, with nearly 10 million deaths in 2020. Advances in genomic technologies have revolutionized cancer research, shifting focus towards precision medicine based on comprehensive tumour genomic profiling. Concurrently, deep learning (DL) has emerged as a powerful paradigm for complex biological data. This review critically assesses recent advances in DL applications for tumour genomics, emphasizing four key domains: DNA sequencing analysis for mutation detection, gene expression profiling for cancer subtype classification, methylation function prediction for epigenetic characterization and integrative multi-omics approaches for comprehensive tumour profiling. We systematically analyse how different DL architectures-including convolutional neural networks, recurrent neural networks, graph neural networks, autoencoders and transformers-address specific challenges in cancer genomics. Our review highlights how these approaches significantly enhance detection sensitivity for genomic alterations, improve cancer subtype stratification, identify novel biomarkers and optimize therapeutic target selection. We examine technical challenges in DL implementation, including model interpretability, data scarcity, computational requirements and integration issues, alongside emerging solutions such as explainable AI, federated learning, and multi-modal frameworks. By synthesizing methodological innovations and identifying research directions, this review provides bioinformaticians and cancer researchers with a roadmap for leveraging DL to advance precision oncology.

References

1. 1.Zhou, Z., Li, Z., Wang, S., Ang, M. Y., Xiao, J., & Choo, S. W. (2026). Deep learning in tumour genomics: from multi-omics integration to precision oncology. Open Biology. https://doi.org/10.1098/rsob.260117