DOI: 10.4103/bbrj.bbrj_72_26 ISSN: 2588-9834

Machine Learning-based Classification of Exercise Status Using miR-21 Expression and Multivariable Participant Profiles

Nana EL Dawy Ahmed Hefny, Fatma Hassan Abd Elbasset Mourgan, Dhamia Ahmed Atiyah, Manaf AlMatar

Background:

Circulating microRNAs (miRNAs) have emerged as promising molecular indicators of exercise-induced physiological adaptation. However, individual miRNAs have shown variable discriminatory performance across exercise contexts. miR-21 is of particular interest because of its proposed involvement in vascular, inflammatory, metabolic, and adaptive responses to exercise.

Methods:

This exploratory observational proof-of-concept study included 20 observations: 10 postexercise (EX) and 10 preexercise/nonexercise (NON). The dataset included age, sex, body mass index, dietary habits, smoking status, alcohol consumption, medical history, exercise type, duration, intensity, frequency, and miR-21 folding values. Univariate analysis of miR-21 folding was performed using Welch’s independent-samples t -test and the Mann–Whitney U -test. Three supervised machine-learning classifiers were applied to the integrated multivariable profile: L2-regularized Logistic Regression, Support Vector Machine with Radial Basis Function (SVM-RBF), and Gaussian Naive Bayes. Model performance was evaluated using repeated stratified 5-fold cross-validation.

Results:

miR-21 folding values alone did not clearly separate EX and NON observations, with substantial overlap between the two groups and nonsignificant univariate results. In contrast, the integrated multivariable profile achieved high internal classification performance across the selected models. Mean accuracy was 99.0% for Logistic Regression, 98.5% for SVM-RBF, and 99.5% for Gaussian Naive Bayes. The mean area under the curve (AUC) values ranged from 0.995 to 1.000, indicating strong internal discriminatory performance within the exploratory dataset.

Conclusions:

The findings suggest that miR-21 should not be interpreted as a standalone classifier of exercise status. Instead, miR-21 may contribute to a broader multivariable exercise-response signature when combined with demographic, lifestyle, clinical, and exercise-related variables.

More from our Archive