Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

IGCN: Integrative Graph Convolution Networks for patient level insights and biomarker discovery in multi-omics integration

Cagri Ozdemir, Mohammad Al Olaimat, Yashu Vashishath, Serdar Bozdag, Alzheimer's Disease Neuroimaging Initiative

TL;DR

IGCN tackles the challenge of explainable multi-omics integration for cancer subtype and biomedical classification by deploying parallel GCNs for each omics type and a personalized attention mechanism to weight modalities at the patient level. It additionally includes a per-sample biomarker ranking module to identify informative features across omics layers, enhancing interpretability and biomarker discovery. Across BRCA, GBM, ROSMAP, and ADNI datasets, IGCN outperforms a broad set of state-of-the-art and baseline methods, demonstrating both superior accuracy and meaningful interpretability. This work advances precision medicine by providing actionable, patient-specific insights into which data types and biomarkers drive predictions, addressing a key gap in multi-omics integration.

Abstract

Developing computational tools for integrative analysis across multiple types of omics data has been of immense importance in cancer molecular biology and precision medicine research. While recent advancements have yielded integrative prediction solutions for multi-omics data, these methods lack a comprehensive and cohesive understanding of the rationale behind their specific predictions. To shed light on personalized medicine and unravel previously unknown characteristics within integrative analysis of multi-omics data, we introduce a novel integrative neural network approach for cancer molecular subtype and biomedical classification applications, named Integrative Graph Convolutional Networks (IGCN). IGCN can identify which types of omics receive more emphasis for each patient to predict a certain class. Additionally, IGCN has the capability to pinpoint significant biomarkers from a range of omics data types. To demonstrate the superiority of IGCN, we compare its performance with other state-of-the-art approaches across different cancer subtype and biomedical classification tasks.

IGCN: Integrative Graph Convolution Networks for patient level insights and biomarker discovery in multi-omics integration

TL;DR

IGCN tackles the challenge of explainable multi-omics integration for cancer subtype and biomedical classification by deploying parallel GCNs for each omics type and a personalized attention mechanism to weight modalities at the patient level. It additionally includes a per-sample biomarker ranking module to identify informative features across omics layers, enhancing interpretability and biomarker discovery. Across BRCA, GBM, ROSMAP, and ADNI datasets, IGCN outperforms a broad set of state-of-the-art and baseline methods, demonstrating both superior accuracy and meaningful interpretability. This work advances precision medicine by providing actionable, patient-specific insights into which data types and biomarkers drive predictions, addressing a key gap in multi-omics integration.

Abstract

Developing computational tools for integrative analysis across multiple types of omics data has been of immense importance in cancer molecular biology and precision medicine research. While recent advancements have yielded integrative prediction solutions for multi-omics data, these methods lack a comprehensive and cohesive understanding of the rationale behind their specific predictions. To shed light on personalized medicine and unravel previously unknown characteristics within integrative analysis of multi-omics data, we introduce a novel integrative neural network approach for cancer molecular subtype and biomedical classification applications, named Integrative Graph Convolutional Networks (IGCN). IGCN can identify which types of omics receive more emphasis for each patient to predict a certain class. Additionally, IGCN has the capability to pinpoint significant biomarkers from a range of omics data types. To demonstrate the superiority of IGCN, we compare its performance with other state-of-the-art approaches across different cancer subtype and biomedical classification tasks.
Paper Structure (14 sections, 15 equations, 3 figures, 3 tables, 1 algorithm)

This paper contains 14 sections, 15 equations, 3 figures, 3 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Results
  3. Overview of IGCN architecture
  4. Biomedical classification tasks
  5. IGCN demonstrated superior performance in various cancer molecular subtype prediction and biomedical classification tasks
  6. Attention-driven Interpretability in IGCN
  7. Ablation study
  8. Discussion
  9. Methods
  10. Customizing GCN for omics-focused learning
  11. Computing attention coefficients and predictions
  12. Uncovering significant biomarkers during the prediction process
  13. Acknowledgements
  14. Code availability

Figures (3)

  • Figure 1: Overview of IGCN architecture on three similarity networks. IGCN employs an integration module to fuse the node embeddings. Simultaneously, IGCN assigns attention coefficients to features for individual samples across diverse omics data types.
  • Figure 2: The boxplots show the distribution of macro F1 scores of ten different runs on (a) TCGA-GBM and (b) ADNI datasets for all methods. The means and standard deviations of these runs are shown in Table \ref{['tab:ex_res']}. Wilcoxon rank-sum test p-values were computed between IGCN and other methods to compare the distribution of box plots, representing p-value $<$ 0.001 by ***, else if $<$ 0.01 by **, and else if $<$ 0.05 by *.
  • Figure 3: Attention coefficients of 50 test samples of (a) TCGA-BRCA and (b) ROSMAP datasets. IGCN provides an attention mechanism, which computes a specific attention coefficient for each node embedding. This speciality might allow us to investigate which feature is most informative for each sample in different node type prediction.