Machine learning algorithms applied to a preclinical COPD model identified neutrophils, TNFα mRNA, CRP, and VO₂ as top predictors of skeletal muscle wasting (Polynomial regression R²=0.81).
Machine learning models successfully identified systemic and pulmonary inflammatory markers as robust predictors of skeletal muscle wasting in a preclinical COPD model.
Effect estimate: R² 0.81
Abstract Rationale Skeletal muscle wasting is a major extrapulmonary complication of chronic obstructive pulmonary disease (COPD), affecting up to 50% of patients and contributing to poor outcomes and increased mortality. Despite its clinical relevance, therapeutic options remain limited due to insufficient evidence supporting predictive biological features of muscle dysfunction. Inflammation—both pulmonary and systemic—has emerged as a potential driver of muscle degradation, independent of physical inactivity. To address this gap, we applied machine learning (ML) algorithms to a preclinical COPD model to identify biological features predictive of muscle wasting, aiming to inform biomarker-guided strategies for early detection and targeted intervention. Methods State-of-the-art ML algorithms were trained, applied to, and validated on previously collected data from our laboratory’s COPD mouse model. Male Balb/c mice were exposed to cigarette smoke (CS) or room air (Sham) for 8 weeks. Twelve biological features were measured, including lung inflammation markers (BALF cell counts, TNFα mRNA, immune cell counts), muscle parameters (Tibialis Anterior weight, specific force, VO₂), and systemic markers (CRP, oxidative stress). After data cleaning and normalization, feature selection was performed using Recursive Feature Elimination and Greedy Feature Selection. Models including Random Forest, XGBoost, and Ridge Regression were trained and validated using 5-fold cross-validation. Results Polynomial regression (degree 3) with 7 selected features achieved a R² of 0.81 and Root Mean Square Error (RMSE) of 1573.03. Random Forest demonstrated the most stable performance (R² = 0.73, %RMSE = 9%). SHAP analysis identified neutrophils, TNFα mRNA, CRP, and VO₂ as top predictors. Inflammation markers correlated positively with each other and negatively with muscle mass and function. Conclusion ML-based feature selection highlights inflammation as a robust predictor of muscle wasting. This approach demonstrates potential for translating preclinical insights into clinical biomarker discovery, enabling data-driven feature selection for early identification and personalised management of muscle dysfunction in COPD patients. This abstract is funded by: NHMRC Project Grant APP1138915
Vlahos et al. (Mon,) conducted a other in Chronic obstructive pulmonary disease (COPD) with skeletal muscle wasting. Cigarette smoke (CS) vs. Room air (Sham) was evaluated on Prediction of muscle wasting (R² and RMSE) (R² 0.81). Machine learning algorithms applied to a preclinical COPD model identified neutrophils, TNFα mRNA, CRP, and VO₂ as top predictors of skeletal muscle wasting (Polynomial regression R²=0.81).
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: