Navigating the disease landscape: knowledge representations for contextualizing molecular signatures

Mansoor Saqi, Artem Lysenko, Yike Guo, Tatsuhiko Tsunoda, Charles AuffrayBriefings in Bioinformatics 2019 · OpenReview

TL;DR

This review explores how to make sense of complex molecular data from diseases by integrating it with existing biological knowledge. The authors highlight three key approaches – pathway-centric, molecular network-centric, and knowledge graphs – that help researchers understand disease mechanisms and pave the way for personalized medicine.

What Problem Does This Paper Address?

Modern medicine generates vast amounts of molecular data (like gene expression profiles) from patients, which can reveal subtypes of diseases. However, these "molecular signatures" alone don't tell us why a disease is happening or how to treat it effectively. The challenge is to contextualize this raw data by integrating it with existing biological knowledge (like known pathways and protein interactions) to gain mechanistic insight into disease processes. This is crucial for developing personalized treatments and understanding complex diseases.

Methods

The paper reviews three main approaches for representing and integrating biological knowledge to contextualize molecular signatures:

  • Pathway-centric approaches: These focus on mapping molecular signatures to known biological pathways (e.g., signaling cascades, metabolic routes). Tools like SBML and BioPAX are used to represent these pathways.
  • Molecular network-centric approaches: These represent biological entities (genes, proteins) as nodes and their interactions as edges in a network. Tools like GeneMANIA and STRING are used to build and analyze these networks, looking for "neighborhoods" of disease-associated genes.
  • Knowledge graph approaches: These represent biological statements (e.g., "protein A interacts with protein B," "gene X is associated with disease Y") as structured graphs. Technologies like RDF and ontologies (e.g., OWL, BEL) are used to build these comprehensive knowledge bases.

The authors emphasize that data integration is a common and crucial component across all these approaches.

Datasets

The paper uses examples from existing biological databases and literature to illustrate the concepts. Specifically, it mentions:

  • Asthma gene signatures: Both a differential expression gene signature from a specific study (GSE43696) and a manually curated core asthma gene signature from the DisGeNET database are used as practical examples.
  • STRING database: Used for protein-protein interaction networks.
  • GeneMANIA: Used for constructing network neighborhoods.
  • Hetionet v1.0: A graph database resource integrating various biomedical information, used to query connections between genes, diseases, pathways, and drugs.
  • Reactome: A pathway database used in conjunction with the Tied Diffusion through Interacting Events method.

No new datasets are introduced in this review paper; it focuses on how existing data can be leveraged.

Key Insights and Ideas

  • Context is King: Molecular signatures are only truly useful when placed within a broader biological context. Without this, they offer limited mechanistic insight.
  • Multiple Levels of Granularity: Disease mechanisms can be understood at different levels, from detailed molecular pathways to broader correlative relationships. Effective representations need to accommodate this.
  • Data Integration is Essential: Regardless of the representation approach, integrating diverse data types (genomics, proteomics, clinical data, literature) is fundamental for a comprehensive understanding.
  • Network and Graph Power: Network and graph-based representations are powerful for integrating, querying, and visualizing complex relationships, making them ideal for biological data analysis.
  • Challenges in Representation: Defining disease mechanisms precisely and representing them in a computable format that can be efficiently queried and used for hypothesis generation remains a significant challenge.
  • The Rise of Knowledge Graphs: Knowledge graphs offer a flexible way to represent complex biological statements and their contexts, enabling sophisticated reasoning and discovery.

Key Results

This paper is a review, not an experimental study, so it doesn't present quantitative results in the traditional sense. However, it illustrates the utility of the discussed approaches through examples:

  • Asthma Network Analysis: By mapping asthma gene signatures to the STRING network, the authors identified potential high-confidence modules and key signaling pathways (HIF, TGF-beta, cytokine-cytokine receptor interaction, circadian rhythm) implicated in asthma.
  • Pathway Identification: Using the Tied Diffusion through Interacting Events method, they identified significant pathways connecting asthma gene sets, highlighting the role of interleukin genes.
  • Knowledge Graph Querying: Using Hetionet, they demonstrated how to query for diseases associated with asthma genes and explore connections between drugs, asthma, and gene signatures, identifying potential therapeutic candidates like niclosamide.

Limitations and Future Work

  • Manual Curation Effort: Constructing rich, information-dense representations often requires significant manual curation, which is time-consuming.
  • Text Mining Limitations: While text mining (TM) is crucial for automating curation, its accuracy and generalizability are still limited.
  • Interoperability: Managing the interoperability between the growing number of different data standards and formalisms remains an open question.
  • Future Directions: The authors suggest that future progress will require more context-aware formats and integration solutions that can handle vast amounts of data and knowledge. They also highlight the need for automated methods to support biomedical innovation, especially given the increasing complexity and cost of drug development. The development of fully mechanistic models of disease through automation is seen as a highly promising emergent strategy.

Summary updated

Abstract

Large amounts of data emerging from experiments in molecular medicine are leading to the identification of molecular signatures associated with disease subtypes. The contextualization of these patterns is important for obtaining mechanistic insight into the aberrant processes associated with a disease, and this typically involves the integration of multiple heterogeneous types of data. In this review, we discuss knowledge representations that can be useful to explore the biological context of molecular signatures, in particular three main approaches, namely, pathway mapping approaches, molecular network centric approaches and approaches that represent biological statements as knowledge graphs. We discuss the utility of each of these paradigms, illustrate how they can be leveraged with selected practical examples and identify ongoing challenges for this field of research.
BibTeX
@article{saqi2019navigating,
  author = {Mansoor Saqi and Artem Lysenko and Yike Guo and Tatsuhiko Tsunoda and Charles Auffray},
  title = {{Navigating the disease landscape: knowledge representations for contextualizing molecular signatures}},
  journal = {Briefings in Bioinformatics},
  year = {2019},
  url = {https://openreview.net/forum?id=D2VBvyGrpm},
}
Figure