MSCFS: inferring circRNA functional similarity based on multiple data sources

Lei Deng, Liang Shu, Cheng Zhou, Xinxu Yuan, Jingpu Zhang, Lei Deng 0002BMC Bioinformatics 2021 · DOI

TL;DR

This paper introduces MSCFS, a novel computational method that integrates multiple data sources to infer the functional similarity between circular RNAs (circRNAs). By combining circRNA-disease associations, circRNA-gene-Gene Ontology (GO) associations, and circRNA sequence information, MSCFS provides a more comprehensive and accurate measure of functional similarity than previous single-source approaches, enabling better understanding of circRNA functions and potential disease associations.

What Problem Does This Paper Address?

Circular RNAs (circRNAs) are a class of non-coding RNAs that have emerged as crucial players in various biological processes and are implicated in numerous human diseases. Identifying their functions is essential for understanding their roles in health and disease. However, inferring the functional similarity between circRNAs is challenging due to the limited availability of comprehensive functional annotations. Existing methods often rely on single data sources, which can be insufficient to capture the complex functional landscape of circRNAs. This paper addresses the need for a robust method that integrates diverse biological data to accurately predict circRNA functional similarity.

Methods

The core of the MSCFS method lies in its multi-source data integration strategy. It generates three distinct circRNA functional similarity networks, each derived from a different data type:

  1. CircRNA-Disease Association Network (CDFS): This network leverages the relationship between circRNAs and diseases. It uses the Directed Acyclic Graph (DAG) structure of disease hierarchies (from the MeSH database) to calculate semantic similarity between diseases. This semantic similarity is then used to infer circRNA functional similarity based on their associated diseases.
  2. CircRNA-Gene-GO Association Network (GOFS): This approach utilizes Gene Ontology (GO) annotations to understand circRNA functions. It constructs a corpus by combining circRNA-gene associations and GO term relationships. The Onto2Vec method is employed to generate vector representations for biological entities, and then the Skip-gram model from Word2Vec is used to learn gene similarity. Finally, circRNA functional similarity is calculated based on the similarity of their associated genes.
  3. CircRNA Sequence Information Network (SQFS): This network captures functional similarity directly from the sequence of circRNAs. It employs the Chaos Game Representation (CGR) to transform circRNA sequences into a grid-based digital matrix. Feature vectors are then extracted from this matrix, and the Pearson correlation coefficient is used to quantify the similarity between these vectors, representing sequence-based functional similarity.

These three similarity matrices (CDFS, GOFS, SQFS) are then integrated using a weighted sum, where the weights (α, β, γ) are determined through parameter selection experiments.

Datasets

The study utilized the following datasets:

  • MeSH (Medical Subject Headings) descriptor: Downloaded from the National Library of Medicine, used for disease hierarchy and relationships.
  • circR2Disease: A benchmark database for experimentally verified circRNA-disease associations. The authors obtained 418 confirmed associations involving 365 circRNAs and 71 diseases.
  • Gene Ontology (GO): Downloaded in OWL format from the Gene Ontology Consortium (GOC) and GO annotations from the GOA Database.
  • circBase: Used to extract circRNA sequence information and circRNA-gene associations.
  • bioDBnet: Provided GO-gene associations.
  • starbase: Used for circRNA-miRNA association data.
  • Peng et al.'s work [42]: Provided circRNA expression profiling data for 2932 circRNAs.

Key Insights and Ideas

  • The power of integration: The study strongly suggests that no single data source is sufficient to accurately capture circRNA functional similarity. Integrating diverse data types (disease, gene-GO, sequence) leads to a more comprehensive and reliable measure.
  • Disease association as a proxy for function: The paper demonstrates that circRNAs associated with similar diseases often share similar functions, providing a valuable avenue for functional inference.
  • Gene-GO links reveal functional roles: By leveraging the rich information in Gene Ontology, the method can infer functional similarity based on the shared functions of genes associated with circRNAs.
  • Sequence holds functional clues: Even at the sequence level, there are discernible patterns that correlate with functional similarity, highlighting the importance of sequence-based analysis.
  • Validation through multiple metrics: The method's effectiveness is validated by its correlation with independent measures like circRNA-miRNA association and circRNA co-expression similarity, lending strong support to its accuracy.

Key Results

  • The MSCFS method showed a positive correlation with circRNA-miRNA association (R = 0.213) and circRNA co-expression similarity (R = 0.8991).
  • Parameter selection experiments identified optimal weights for the three data sources (α=0.1, β=0.3, γ=0.6).
  • Case studies using the CircFunBase database demonstrated that circRNAs with high MSCFS similarity scores were indeed related to the same diseases (e.g., gastric cancer, hypertension), validating the method's predictive power.
  • The study found that circRNA functional similarity obtained by MSCFS is positively correlated with circRNA co-expression similarity (R = 0.8991).

Limitations and Future Work

  • Data quality bias: The authors acknowledge that the integrated results may be "somewhat biased due to the quality of some data sources." This is a common challenge in bioinformatics, where data completeness and accuracy can vary.
  • Limited functional annotations: The paper highlights the ongoing challenge of insufficient functional annotations for circRNAs in public databases, which can limit the accuracy of any functional inference method.

For future work, the authors suggest integrating "more and more reliable data to further improve the accuracy of circRNA functional similarity calculations." This implies a continuous effort to expand the data sources and refine the integration strategies as new and higher-quality biological data become available.

Summary updated

BibTeX
@article{deng2021mscfs,
  author = {Lei Deng and Liang Shu and Cheng Zhou and Xinxu Yuan and Jingpu Zhang and Lei Deng},
  title = {{MSCFS: inferring circRNA functional similarity based on multiple data sources}},
  journal = {BMC Bioinformatics},
  year = {2021},
  doi = {10.1186/s12859-021-04287-1},
}
Figure