Integrating Proteomic, Metabolomic, and Genomic Numerical Biomarkers via Graph Neural Networks for Early-Stage Pancreatic Cancer Stratification

• Sunday Sunday

  Ladoke Akintola University of Technology

  Author

Pancreatic ductal adenocarcinoma (PDAC) remains one of the most lethal malignancies, with a 5-year survival rate below 13% primarily due to late-stage diagnosis and the absence of reliable early detection biomarkers . While multi-omics approaches have shown promise in cancer subtyping, current integration methods often treat omics layers as isolated data streams, failing to capture the complex inter-omics dependencies that characterize early carcinogenesis . This study proposes a novel Graph Neural Network (GNN)-based framework that integrates numerical biomarkers from proteomic, metabolomic, and genomic data to enable early-stage PDAC stratification. Using publicly available TCGA and pre-diagnostic cohort datasets, we constructed patient similarity networks using Mahalanobis distance and density-based methods, implemented multi-view attention mechanisms to fuse complementary information across omics layers, and employed multi-task learning via cross-omics tensors. The proposed framework achieved an overall classification accuracy of 89.4% (AUC = 0.93) for distinguishing Stage I/II PDAC from high-risk controls, significantly outperforming traditional machine learning baselines (p < 0.001) and CA19-9 alone (sensitivity 83.3% vs. 79.0%). Feature importance analysis identified MUC4, KRAS mutation status, and a 4-metabolite signature as the top predictive biomarkers. This framework provides a replicable computational approach for early cancer detection, with implications for precision screening programs and the development of non-invasive diagnostic tools.

• PDF

Copyright (c) 2026 Sunday Sunday (Author)

This work is licensed under a Creative Commons Attribution 4.0 International License.