DMirNet: Inferring direct microRNA-mRNA association networks

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• MicroRNAs (miRNAs) regulate biological processes by affecting target messenger RNAs (mRNAs).
• Existing computational methods for inferring miRNA-mRNA networks from expression profiles suffer from erroneous links, unreliable correlation estimates due to high-dimensional data, and inconsistent performance across datasets.
• A robust and reliable method is needed to accurately identify direct miRNA-mRNA associations.

Purpose of the Study:

• To develop a novel framework, DMirNet, for identifying direct miRNA-mRNA association networks.
• To address limitations of conventional methods in inferring accurate and robust miRNA-mRNA associations.
• To provide a reliable model that performs well across different datasets.

Main Methods:

• DMirNet integrates three direct correlation estimation methods: Corpcor, SPACE, and Network deconvolution.
• The bootstrapping method is employed to effectively utilize limited expression profile data.
• A rank-based Ensemble aggregation strategy is used to ensure model reliability and robustness across datasets.

Main Results:

• DMirNet successfully infers direct miRNA-mRNA association networks.
• Empirical experiments demonstrate the effectiveness of DMirNet's components and its superior performance compared to state-of-the-art ensemble models (up to 1.29x improvement).
• The study identified 43 novel multi-cancer-related miRNA-mRNA associations from the top 1000 inferred direct associations.

Conclusions:

• DMirNet is a promising tool for discovering novel direct miRNA-mRNA relationships and elucidating complex association networks.
• The framework's ability to infer direct relationships aids in reconstructing various direct regulatory pathways.
• DMirNet offers a reliable approach for advancing the understanding of miRNA-mediated gene regulation.