Machine Learning Based Methods and Best Practices of microRNA-Target Prediction and Validation

Area of Science:

• Molecular Biology
• Genomics
• Bioinformatics

Background:

• Noncoding RNAs (ncRNAs), particularly microRNAs (miRNAs), are key regulators of gene expression.
• Dysregulated miRNAs are implicated in various diseases, including cancer, acting as oncogenic or tumor-suppressive molecules.
• miRNA-based therapeutics are advancing in clinical trials for cancer treatment.

Purpose of the Study:

• To highlight the importance of accurate miRNA and mRNA target prediction for effective cancer therapeutics.
• To discuss essential requirements and best practices for developing and validating miRNA target prediction tools.
• To explore the application of these predictions in cancer diagnosis and therapy.

Main Methods:

• Review of existing approaches for miRNA and miRNA target prediction based on sequence and structure features.
• Emphasis on machine learning (ML) methods for training, testing, and validation using high-throughput sequencing data.
• Identification of common features used in miRNA-binding site prediction for ML models.

Main Results:

• The abundance of sequencing data and databases facilitates ML-driven prediction and validation.
• ML approaches offer improved accuracy in predicting miRNA-mRNA interactions.
• Established best practices enhance the reliability of miRNA target predictions.

Conclusions:

• Accurate miRNA target prediction is vital for minimizing side-effects and maximizing efficacy in miRNA-based cancer therapies.
• Machine learning methods, leveraging diverse features and data, are instrumental in advancing prediction accuracy.
• Improved prediction tools hold significant potential for enhancing cancer diagnosis and therapeutic strategies.