Proteins at the intersection of circadian rhythms and metabolic dysfunction-associated steatotic liver disease: an 18-protein panel as a novel predictive biomarker set

Disruption of circadian rhythmicity is highly prevalent in modern society and contributes to the epidemic of metabolic disorders. However, the relationship between circadian rhythm disorder and Metabolic Dysfunction-Associated Steatotic Liver (MASLD) remained poorly elucidated. Relative amplitude (RA), a metric quantifying the degree of disruption in rest-activity circadian rhythms, was calculated based on accelerometry data from UK Biobank. Basic characteristic analysis and multivariable logistic regression was used to analyze the association between RA and MASLD. Mediation analysis, functional enrichment analysis (Kyoto Encyclopedia of Genes and Genomes, KEGG) and Protein-Protein Interaction (PPI) network analysis and multiple machine learning algorithms, including Random Forest, XGBoost, logistic regression, and Support Vector Machine (SVM), were employed to identify potential protein biomarkers and construct a predictive model for RA-related MASLD risk assessment. Among 81,430 UK Biobank participants with valid accelerometry, RA was lower in individuals with MASLD versus those without (P < 0.001). Lower RA was associated to higher prevalence of MASLD (crude OR = 2.61; 95% CI [2.42, 2.81]; P < 0.001), and the association remained significant in a fully adjusted model (adjusted OR = 1.15; 95% CI [1.02,1.31]; P = 0.026), demonstrating RA as a factor independently associated with MASLD. Furthermore, 18 candidate proteins were identified as potential biomarkers for predicting RA-related MASLD. The 18-protein model demonstrated excellent predictive performance across multiple machine learning methods, with high Area Under the Curve (AUC) values in Receiver Operating Characteristic (ROC) analysis: Random Forest (AUC = 0.944), XGBoost (AUC = 0.946), Logistic Regression (AUC = 0.946), and SVM (AUC = 0.947). The model also exhibited superior discriminatory ability in predictive probability distribution, indicating strong predictive potential. Additionally, an online predictive tool based on this model has been contributed. Lower RA is independently associated with MASLD. We highlight 18 overlapping plasma proteins linked to both RA and MASLD as potential biomarkers for predicting RA-associated MASLD and as candidate therapeutic targets.