Machine Learning-Based Ensemble Feature Selection and Nested Cross-Validation for miRNA Biomarker Discovery in Usher Syndrome.

Usher syndrome (USH) is a rare genetic disorder affecting vision, hearing, and balance. Identifying reliable biomarkers is crucial for early diagnosis and understanding disease mechanisms. MicroRNAs (miRNAs), key regulators of gene expression, hold promise as biomarkers for USH. This study aimed to identify a minimal subset of miRNAs that could serve as biomarkers to effectively differentiate USH from controls. We employed ensemble feature selection techniques to select the top miRNAs appearing in at least three algorithms. Machine learning models were trained and tested using this subset, followed by validation on an independent 10% sample. Our approach identified 10 key miRNAs as potential biomarkers for USH. To further validate their biological relevance, we conducted pathway analysis, which revealed significant pathways associated with USH. Furthermore, our approach achieved high classification performance, with an accuracy of 97.7%, sensitivity of 98%, specificity of 92.5%, F1 score of 95.8%, and an AUC of 97.5%. These findings demonstrate that combining ensemble feature selection with machine learning provides a robust strategy for miRNA biomarker discovery, advancing USH diagnosis and molecular understanding.