Heterogeneous graph inference with range constrainted L2,1-collaborative matrix factorization for small molecule-miRNA association prediction | JoVE Visualize

Area of Science:

Computational Biology and Bioinformatics
Genomics and Molecular Biology
Pharmacology and Drug Discovery

Background:

MicroRNAs (miRNAs) are crucial regulators of gene expression and biological processes, making them key targets for small molecule (SM) drugs in disease therapy.
Predicting SM-miRNA associations is vital for drug discovery, but existing methods struggle with sparse association networks and imprecise similarity metrics.
Heterogeneous graph inference is a common approach, but its effectiveness is limited by data sparsity and inaccurate similarity calculations.

Purpose of the Study:

To develop an advanced computational method, HGIRCLMF, for accurately predicting potential small molecule-microRNA associations.
To address the limitations of sparsity and imprecise similarity metrics in existing SM-miRNA association prediction models.
To enhance the reliability and accuracy of identifying novel therapeutic targets through improved association inference.

Main Methods:

Computed multi-source similarities for small molecules (SMs) and miRNAs, integrating them into comprehensive similarity metrics.
Employed a novel range constrained L2,1-collaborative matrix factorization (RCLMF) model to address matrix sparsity and enhance SM-miRNA edge robustness.
Constructed a heterogeneous network using processed biological data and applied the HGIRCLMF model for inferring unknown association scores.

Main Results:

The proposed HGIRCLMF method achieved superior performance in predicting SM-miRNA associations, demonstrated by the highest areas under the curve in cross-validation experiments.
HGIRCLMF outperformed six state-of-the-art computational approaches on two distinct datasets, indicating its enhanced predictive accuracy.
Case studies confirmed the practical applicability and predictive power of HGIRCLMF in real-world drug discovery scenarios.

Conclusions:

HGIRCLMF effectively overcomes the challenges of data sparsity and inaccurate similarity metrics in SM-miRNA association prediction.
The method provides a robust and accurate computational tool for identifying potential SM-miRNA associations, aiding in drug target discovery.
This research contributes a significant advancement to the field of bioinformatics and computational drug discovery.