A network-based pathway-extending approach using DNA methylation and gene expression data to identify altered pathways

Jie Li, Qiaosheng Zhang, Zhuo Chen, Dechen Xu, Yadong WangScientific 2019 · DOI

TL;DR

This paper introduces a novel network-based approach that integrates DNA methylation and gene expression data to "extend" existing biological pathways. By doing so, it aims to uncover new pathway components and better identify cancer-related pathways, outperforming traditional methods in identifying significant pathways across multiple cancer datasets.

What Problem Does This Paper Address?

Understanding the complex molecular mechanisms driving cancer is crucial for developing effective therapies. While pathway analysis is a powerful tool for this, traditional methods often struggle to fully leverage the wealth of high-throughput multi-omics data and large-scale biological networks available today. Specifically, existing approaches for pathway analysis can be limited in their ability to:

  • Integrate diverse data types: Effectively combining information from different molecular layers like DNA methylation and gene expression.
  • Expand pathway knowledge: Identifying novel genes or interactions that contribute to a pathway's function, beyond its currently defined components.
  • Capture complex interactions: Recognizing the intricate crosstalk and interdependencies between different biological pathways.

This paper aims to bridge these gaps by proposing a method that uses network structures and multi-omics data to enrich and extend existing pathways, thereby improving their relevance and power in cancer research.

Methods

The core of this work is a novel network-based approach that extends original pathways by integrating DNA methylation and gene expression data. Here's a breakdown of the key steps:

  1. Weighted Gene-Gene Interaction Network Construction:
    • The authors start with a Protein-Protein Interaction (PPI) network as a foundational structure.
  2. Crucially, they assign edge weights to this network by integrating DNA methylation and gene expression data. This is achieved using Principal Component Analysis (PCA) for dimensionality reduction of CpG sites in methylation data, followed by Sparse Canonical Correlation Analysis (SCCA) to quantify the relationship between the reduced methylation data and gene expression data for gene pairs. This weighted network reflects the biological state (e.g., cancer vs. normal tissue).
  3. This process is performed separately for different phenotypes (e.g., cancer tissue and normal tissue) to create phenotype-specific weighted networks.
  4. Pathway Extension:
    • Original pathways are then "extended" within these weighted phenotype-specific networks.
  5. The method uses a limited k-Walks algorithm to identify neighboring genes that are highly relevant to the genes within an original pathway. These neighboring genes are considered new components of the extended pathway.
  6. The extended pathways from different phenotypes are then merged into a single, comprehensive extended pathway gene list.
  7. Cancer-Related Pathway Identification:
    • The gene lists of these extended pathways are then fed into classical gene set analysis methods:
  8. Over-Representation Analysis (ORA): Here referred to as EP-ORA (Extended Pathway ORA).
  9. Functional Class Scoring (FCS): Specifically, Gene Set Enrichment Analysis (GSEA), referred to as EP-GSEA (Extended Pathway GSEA).
  10. This allows for the identification of significantly altered pathways that are correlated with cancer.

Datasets

The study utilizes publicly available datasets from The Cancer Genome Atlas (TCGA). Specifically, they analyzed data from three cancer types:

  • BRCA (Breast Invasive Carcinoma): 33 cancer samples and 37 normal tissue samples.
  • LUAD (Lung Adenocarcinoma): 69 samples (49 cancer, 20 normal).
  • COAD (Colon Adenocarcinoma): 42 samples (26 cancer, 16 normal).

All methylation data were from Illumina Human Methylation 450k Chip, and gene expression data were from Agilent G4502A or Illumina HiSeq platforms. Gene sets for biological pathways were sourced from the ConsensusPathDB website, with 281 KEGG pathways used for analysis.

Key Insights and Ideas

  • Beyond Static Pathways: The core idea is that pathways are not static entities but dynamic and interconnected systems. Extending them with network information and multi-omics data provides a more realistic and comprehensive view.
  • DNA Methylation as a Regulator: The integration of DNA methylation data is key, as it's known to influence gene transcription. This allows the weighted networks to capture regulatory effects beyond just gene expression.
  • Network Topology Matters: By using weighted networks, the method accounts for the strength of interactions, not just their presence, which is a significant improvement over previous network extension methods that ignored edge weights.
  • Discovering Hidden Components: The extension process effectively identifies "neighboring" genes that play crucial roles in pathway regulation but might not be explicitly listed in canonical pathway databases.
  • Crosstalk Revelation: The extended pathways can better reveal the complex crosstalk between different biological pathways, which is fundamental to understanding complex diseases like cancer.

Key Results

  • Pathway Extension Effectiveness: The extension process significantly increased the size of original pathways, suggesting the inclusion of many new, relevant genes. For example, in BRCA, pathway sizes increased by an average of 28.30% to 224.56%.
  • Improved Cancer Pathway Identification:
    • In BRCA, EP-ORA identified 18 significant pathways compared to 6 by standard ORA. Several cancer-related pathways (e.g., p53 signaling, DNA replication) were uniquely identified by EP-ORA.
  • EP-GSEA identified 3 significant pathways in BRCA, whereas standard GSEA found none, highlighting the benefit of extended pathways even with low-power GSEA.
  • Analyses on LUAD and COAD datasets also showed that EP-ORA and EP-GSEA consistently identified more cancer-relevant pathways, often ranking them higher than standard ORA and GSEA. For instance, the "bile secretion pathway" was uniquely identified by EP-ORA in LUAD.
  • Validation of Identified Pathways: The identified pathways were frequently confirmed by existing literature to be associated with the respective cancers, lending strong support to the method's validity.

Limitations and Future Work

  • Computational Cost: While not explicitly detailed as a limitation, the integration of multi-omics data and network analysis can be computationally intensive.
  • Parameter Optimization: The method involves parameters (e.g., c1, c2 in the weight calculation) that might require careful tuning for optimal performance across different datasets.
  • Future Work: The authors suggest employing more datasets to further assess the method's validity. They also provide their code for broader use, indicating a commitment to further development and application. The potential to explore more complex network structures and data types remains an avenue for future research.

Summary updated

Abstract

Pathway analysis allows us to gain insights into a comprehensive understanding of the molecular mechanisms underlying cancers. Currently, high-throughput multi-omics data and various types of large-scale biological networks enable us to identify cancer-related pathways by comprehensively analyzing these data. Combining information from multidimensional data, pathway databases and interaction networks is a promising strategy to identify cancer-related pathways. Here we present a novel network-based approach for integrative analysis of DNA methylation and gene expression data to extend original pathways. The results show that the extension of original pathways can provide a basis for discovering new components of the original pathway and understanding the crosstalk between pathways in a large-scale biological network. By inputting the gene lists of the extended pathways into the classical gene set analysis (ORA and FCS), we effectively identified the altered pathways which are correlated well with the corresponding cancer. The method is evaluated on three datasets retrieved from TCGA (BRCA, LUAD and COAD). The results show that the integration of DNA methylation and gene expression data through a network of known gene interactions is effective in identifying altered pathways.
BibTeX
@inproceedings{li2019network,
  author = {Jie Li and Qiaosheng Zhang and Zhuo Chen and Dechen Xu and Yadong Wang},
  title = {{A network-based pathway-extending approach using DNA methylation and gene expression data to identify altered pathways}},
  booktitle = {Scientific},
  year = {2019},
  doi = {10.1038/s41598-019-48372-1},
}
Figure