TL;DR
This paper introduces iCTNet, a Cytoscape plugin that integrates diverse biological data sources (genetics, proteins, tissues, drugs) to build and analyze complex trait networks. This tool aims to bridge the gap between the explosion of biological data and our ability to meaningfully integrate and interpret it, facilitating the discovery of novel disease relationships and therapeutic strategies.
What Problem Does This Paper Address?
The rapid accumulation of biological data from various sources (genomics, proteomics, clinical data) has outpaced our ability to integrate and analyze it effectively. These disparate datasets, while rich in information, often remain disconnected, hindering a holistic understanding of complex human traits and diseases. Researchers face the challenge of connecting molecular-level data with clinical phenotypes, making it difficult to identify underlying mechanisms, discover disease similarities, or propose novel treatments. This paper addresses this by providing a unified platform to integrate and visualize these multi-layered biological networks.
Methods
iCTNet is implemented as a Java plugin for Cytoscape, a popular open-source platform for network analysis and visualization. The core of iCTNet lies in its ability to integrate data from multiple sources to construct multi-partite networks with up to five types of nodes (phenotype/SNP, protein/gene, tissue, drug) and five types of edges.
Key methodological contributions include:
- Data Integration: iCTNet consolidates data from various sources, including GWAS catalogs, protein-protein interaction databases (HPRD), tissue expression data, and drug databases (DrugBank).
- Network Construction: It allows the creation of networks with up to five layers, connecting phenotypes to SNPs, proteins to proteins, diseases to tissues, tissues to genes, and drugs to genes.
- Candidate Gene Prioritization: iCTNet incorporates two algorithms, "random walk with restarts" and "PRINCE," to prioritize candidate genes. These algorithms leverage network topology and association p-values to identify genes that are potentially involved in a disease, even if their direct association is modest.
- Network Visualization and Analysis: The plugin provides a user-friendly interface to search, integrate, visualize, and analyze these complex networks. It offers options for creating general or specific network views, filtering by significance thresholds, and exploring relationships at different degrees of separation.
Datasets
iCTNet integrates data from several publicly available sources:
- Phenotype-SNP Associations: Data from published GWAS articles and the GWAS Catalog, mapped to human genome build 37. This includes over 6,000 trait-SNP associations at a significance level of 10⁻⁴ and over 2,000 at 10⁻⁸.
- Protein Interactions: Over 33,000 interactions among 12,000 proteins from the Human Protein Reference Database (HPRD).
- Tissue Information: Gene expression data from HPRD, linking genes to tissues. This was curated into 38 categories.
- Drug-Gene Interactions: Information from DrugBank v.2.5, linking drugs to their gene targets.
The paper does not introduce new benchmark datasets but rather leverages existing, large-scale biological databases.
Key Insights and Ideas
- The Power of Multi-Partite Networks: The paper highlights that integrating multiple layers of biological information (genetics, proteins, tissues, drugs) into a single network provides a more comprehensive view of complex traits than single-layer networks.
- Bridging Molecular and Clinical Data: iCTNet effectively bridges the gap between molecular-level data (genes, proteins) and clinical phenotypes (diseases, traits), enabling a systems-level understanding.
- Discovering Hidden Relationships: By visualizing and analyzing these integrated networks, researchers can uncover novel genetic similarities between diseases and identify potential drug repurposing opportunities. For example, the visualization of autoimmune diseases revealed close relationships between RA, MS, and T1D, consistent with known HLA associations.
- Prioritizing Research Efforts: The candidate gene prioritization algorithms are crucial for focusing experimental efforts on the most promising genes, saving time and resources.
Key Results
- iCTNet allows the creation of networks with up to five layers of omics information.
- The plugin facilitates the generation of network views for over 200 diseases.
- The candidate gene prioritization algorithms (random walk and PRINCE) are implemented and available for use.
- Qualitative results demonstrate the utility of iCTNet in visualizing relationships among autoimmune diseases, highlighting shared genetic associations and potential drug targets. For instance, a network of five common autoimmune diseases was generated, showing disease similarities and drug-target interactions.
Limitations and Future Work
- Algorithm Recommendation: The authors note that a comprehensive comparison of the prioritization algorithms under extensive parameter ranges has not been performed, making it difficult to recommend a specific algorithm. They encourage users to test both.
- Future Data Integration: The authors plan to integrate additional data sources, including quantitative omics data and miRNA targets.
- Advanced Analysis: Future work may involve developing more advanced analysis tools within the plugin.
- Monthly Updates: The GWAS catalog will be updated monthly to maintain the currency of the data.
Summary updated