Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Graph Representation Learning Strategies for Omics Data: A Case Study on Parkinson's Disease

Elisa Gómez de Lope, Saurabh Deshpande, Ramón Viñas Torné, Pietro Liò, Enrico Glaab, Stéphane P. A. Bordas

TL;DR

This study tackles the challenge of analyzing high-dimensional, noisy, and heterogeneous omics data in Parkinson's disease by evaluating graph representation learning across two cohorts: transcriptomics from PPMI and metabolomics from LUXPARK. It contrasts sample-similarity networks (SSN) and molecular-interaction networks (MINs) built from PPI (STRING) and MMI (STITCH), using a suite of GNNs, graph transformers, and a baseline MLP, with a two-layer default configuration. The SSN adjacency is defined by pairwise cosine similarity with a threshold, A_{ij} = { cs(x_i, x_j) if i ≠ j and cs(x_i, x_j) ≥ s, 0 otherwise }, enabling effective information propagation and interpretability via GNN-Explainer. Key findings show SSNs generally outperform MINs, graph transformers outperform traditional GNNs, and LASSO feature selection enhances performance; biologically, signals converge on the mitochondrial fatty acid oxidation pathway involving SLC25A20, CPT1A, and glutarylcarnitine, offering PD-relevant insights with potential for biomarker discovery.

Abstract

Omics data analysis is crucial for studying complex diseases, but its high dimensionality and heterogeneity challenge classical statistical and machine learning methods. Graph neural networks have emerged as promising alternatives, yet the optimal strategies for their design and optimization in real-world biomedical challenges remain unclear. This study evaluates various graph representation learning models for case-control classification using high-throughput biological data from Parkinson's disease and control samples. We compare topologies derived from sample similarity networks and molecular interaction networks, including protein-protein and metabolite-metabolite interactions (PPI, MMI). Graph Convolutional Network (GCNs), Chebyshev spectral graph convolution (ChebyNet), and Graph Attention Network (GAT), are evaluated alongside advanced architectures like graph transformers, the graph U-net, and simpler models like multilayer perceptron (MLP). These models are systematically applied to transcriptomics and metabolomics data independently. Our comparative analysis highlights the benefits and limitations of various architectures in extracting patterns from omics data, paving the way for more accurate and interpretable models in biomedical research.

Graph Representation Learning Strategies for Omics Data: A Case Study on Parkinson's Disease

TL;DR

This study tackles the challenge of analyzing high-dimensional, noisy, and heterogeneous omics data in Parkinson's disease by evaluating graph representation learning across two cohorts: transcriptomics from PPMI and metabolomics from LUXPARK. It contrasts sample-similarity networks (SSN) and molecular-interaction networks (MINs) built from PPI (STRING) and MMI (STITCH), using a suite of GNNs, graph transformers, and a baseline MLP, with a two-layer default configuration. The SSN adjacency is defined by pairwise cosine similarity with a threshold, A_{ij} = { cs(x_i, x_j) if i ≠ j and cs(x_i, x_j) ≥ s, 0 otherwise }, enabling effective information propagation and interpretability via GNN-Explainer. Key findings show SSNs generally outperform MINs, graph transformers outperform traditional GNNs, and LASSO feature selection enhances performance; biologically, signals converge on the mitochondrial fatty acid oxidation pathway involving SLC25A20, CPT1A, and glutarylcarnitine, offering PD-relevant insights with potential for biomarker discovery.

Abstract

Omics data analysis is crucial for studying complex diseases, but its high dimensionality and heterogeneity challenge classical statistical and machine learning methods. Graph neural networks have emerged as promising alternatives, yet the optimal strategies for their design and optimization in real-world biomedical challenges remain unclear. This study evaluates various graph representation learning models for case-control classification using high-throughput biological data from Parkinson's disease and control samples. We compare topologies derived from sample similarity networks and molecular interaction networks, including protein-protein and metabolite-metabolite interactions (PPI, MMI). Graph Convolutional Network (GCNs), Chebyshev spectral graph convolution (ChebyNet), and Graph Attention Network (GAT), are evaluated alongside advanced architectures like graph transformers, the graph U-net, and simpler models like multilayer perceptron (MLP). These models are systematically applied to transcriptomics and metabolomics data independently. Our comparative analysis highlights the benefits and limitations of various architectures in extracting patterns from omics data, paving the way for more accurate and interpretable models in biomedical research.
Paper Structure (16 sections, 2 equations, 2 figures, 2 tables)

This paper contains 16 sections, 2 equations, 2 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Materials and Methods
  3. Datasets
  4. Modelling Sample Similarity Networks
  5. Modelling Molecular Interactions Networks
  6. Comparative analysis and evaluation
  7. Results and discussion
  8. Sample similarity versus molecular interaction networks
  9. The power of the graph structure: MLP vs GNNs
  10. The effect of feature selection and depth of the network
  11. Transformers generally outperform traditional GNNs
  12. Inductive biases: similarity-based versus random edges
  13. Biological Insights
  14. Future work
  15. Data pre-processing
  16. ...and 1 more sections

Figures (2)

  • Figure 1: Schema of graph representation learning using sample-sample similarity networks modeling pipeline.
  • Figure 2: Schema of graph representation learning using molecular interaction networks modeling pipeline.