ComiR: combinatorial microRNA target prediction tool

Claudia Coronnello, Panayiotis V. BenosNucleic Acids Res 2013 · DOI

TL;DR

ComiR is a novel web tool that predicts how sets of microRNAs (miRNAs) regulate messenger RNAs (mRNAs). It addresses the limitation of existing tools by integrating miRNA expression levels and combinatorial binding effects, leading to more accurate and context-aware target predictions.

What Problem Does This Paper Address?

MicroRNAs (miRNAs) are crucial regulators of gene expression, primarily by binding to the 3' untranslated regions (3'UTRs) of messenger RNAs (mRNAs). While many computational tools exist to predict miRNA-mRNA interactions, they often focus on individual miRNA-target site pairings and overlook crucial biological factors. These factors include:

  • miRNA Expression Levels: The abundance of a miRNA in a cell significantly influences its regulatory capacity.
  • Combinatorial Binding: Multiple miRNAs can bind to the same mRNA, and their combined effect can be more than the sum of individual interactions.
  • System-wide Factors: Competition for binding sites and the presence of other regulatory RNAs can also impact miRNA efficacy.

Existing tools typically predict only a fraction of known miRNA targets, and their predictions often show poor overlap. This suggests a need for more sophisticated approaches that account for the complex, system-wide nature of miRNA regulation. ComiR aims to fill this gap by considering both miRNA expression levels and the combinatorial effects of multiple miRNAs targeting a single mRNA.

Methods

ComiR employs a multi-step approach to predict combinatorial miRNA targeting:

  1. Individual Target Scoring: For each potential miRNA-mRNA interaction, ComiR utilizes four established scoring methods:
  2. miRanda: Adapts the Fermi-Dirac equation to incorporate miRNA expression and sum probabilities over all targets on an mRNA.
  3. PITA: Similar to miRanda, it uses the Fermi-Dirac equation but substitutes standard energies to account for miRNA expression.
  4. TargetScan: Uses the number of seed sequence matches, weighted by miRNA expression.
  5. mirSVR: Uses the mirSVR prediction score, also weighted by miRNA expression.
  6. Score Combination and SVM Integration: The scores from these four methods are then combined using a Support Vector Machine (SVM). This SVM was trained on a high-quality dataset derived from Drosophila AGO1 immunoprecipitation (IP) experiments. A normalization step is applied to make the model applicable to other species.
  7. Combinatorial Scoring: The SVM output provides a probability score for each mRNA being regulated by the set of input miRNAs, considering their expression levels. This score reflects the overall regulatory potential of the miRNA set on a given mRNA.

The tool offers a user-friendly interface for inputting miRNA expression data and selecting species or custom mRNA sequences. It also includes options for filtering miRNAs based on expression levels and considering miRNAs as a set or individually.

Datasets

  • mRNA 3'UTR Sequences: Downloaded from the ENSEMBL Web site (as of September 2012) for Homo sapiens, Mus musculus, Drosophila melanogaster, and Caenorhabditis elegans. For genes with multiple 3'UTRs, the longest sequence was used.
  • miRNA Mature Sequences: Downloaded from miRBase (Release 19).
  • Drosophila AGO1 IP Data: Used to train the SVM model for combining the four primary scoring methods.

Key Insights and Ideas

  • Expression Matters: Integrating miRNA expression levels is crucial for accurate target prediction, as it directly influences the regulatory capacity of a miRNA.
  • Combinatorial Effects are Key: Simply summing individual miRNA-target predictions is insufficient. ComiR's approach of combining scores from multiple methods and using an SVM captures synergistic or antagonistic effects of multiple miRNAs.
  • Cross-Species Applicability: The SVM model, trained on Drosophila data and normalized, demonstrates effectiveness in predicting miRNA targets in other species, highlighting the generalizability of the underlying principles.
  • Beyond Simple Binding: ComiR moves beyond just identifying potential binding sites to assessing the regulatory potential of a miRNA set, which is a more biologically relevant measure.

Key Results

  • ComiR demonstrates a "significant improvement over standard miRNA prediction algorithms adapted for combinatorial miRNA targeting." (While specific quantitative metrics like AUC or precision/recall are not explicitly detailed in the abstract or introduction, the paper emphasizes this improvement).
  • The tool provides a ranked list of genes based on their predicted regulatory probability by a given miRNA set.
  • The output can be visualized and filtered, allowing users to identify the most relevant targets and compare results from different miRNA expression inputs.
  • The computational time for analysis is relatively fast, ranging from seconds to minutes for typical inputs, with custom mRNA sequences linearly increasing the time.

Limitations and Future Work

  • Unsolved Issues: The authors acknowledge that several aspects of miRNA regulation remain unsolved, including:
  • Competition for Ago proteins: The extent to which miRNAs compete for binding to Argonaute (Ago) proteins is not fully accounted for.
  • Cooperative effects of miRNAs: The precise mechanisms and extent of cooperative interactions between miRNAs are still being investigated.
  • Future Directions: The authors plan to extend the miRNA binding models to incorporate these more complex effects as more relevant data become available. They also mention the potential to evaluate the effect of single-nucleotide polymorphisms (SNPs) on miRNA regulation by comparing ComiR scores for wild-type and SNP sequences.

Summary updated

BibTeX
@article{coronnello2013comir,
  author = {Claudia Coronnello and Panayiotis V. Benos},
  title = {{ComiR: combinatorial microRNA target prediction tool}},
  journal = {Nucleic Acids Res.},
  year = {2013},
  doi = {10.1093/nar/gkt379},
}
Figure