Updated review of advances in microRNAs and complex diseases: towards systematic evaluation of computational models

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Established evaluation strategies for microRNA-disease association (MDA) prediction models are lacking.
• Current model evaluation relies on researcher-dependent choices for K-fold cross-validation parameters, metrics, and study designs.
• Inconsistency in evaluation hinders fair comparison and systematic assessment of predictive performance.

Approach:

• A comprehensive analysis of evaluation methodologies employed by 29 state-of-the-art MDA prediction models was conducted.
• The study systematically reviewed practices including K-fold cross-validation, case studies, parameter sensitivity tests, and ablation studies.
• The analysis focused on identifying commonalities, variations, and potential areas for standardization.

Key Points:

• Evaluation metrics and procedures for MDA models vary significantly across studies.
• Parameters like K in cross-validation, query disease selection, and additional tests are not standardized.
• A lack of standardized evaluation limits the reproducibility and comparability of MDA prediction models.

Conclusions:

• A need exists for a universally accepted framework for evaluating computational MDA models.
• The proposed evaluation workflow aims to provide a systematic and fair method for assessing predictive performance.
• Implementing this workflow will enhance the reliability and comparability of future MDA prediction models.