Machine learning approaches for predicting the small molecule-miRNA associations: a comprehensive review

Area of Science:

• Computational Biology and Bioinformatics
• Drug Discovery and Development

Background:

• MicroRNAs (miRNAs) are key regulators implicated in human diseases.
• Small molecules (SMs) can modulate miRNA activity, offering therapeutic potential.
• Experimental determination of SM-miRNA associations (SMAs) is challenging due to miRNA structural complexity.

Purpose of the Study:

• To review machine learning (ML) approaches for predicting SMAs.
• To provide a comprehensive overview of data resources, feature extraction, and ML algorithms.
• To analyze trends and performance of existing ML-based SMA prediction methods.

Main Methods:

• Compilation and analysis of 32 distinct ML-based SMA prediction methods.
• Classification of methods into six categories: classical ML, deep learning, matrix factorization, network propagation, graph learning, and ensemble learning.
• Review of data resources, feature engineering techniques, and performance metrics for ML algorithms.

Main Results:

• Detailed census and categorization of 32 ML-based SMA prediction methodologies.
• Analysis of trends in algorithm usage and performance across different ML approaches.
• Identification of key principles and comparative performance of state-of-the-art methods.

Conclusions:

• Machine learning offers a cost-effective and efficient alternative to experimental screening for SMAs.
• This review provides critical insights into current ML methodologies for SMA prediction.
• Highlights key areas for future research to advance ML-driven drug discovery targeting miRNAs.