Machine learning-based discrimination of metabolic dysfunction-associated steatotic liver disease: implications of tryptophan metabolites as potential biomarkers.

OBJECTIVES: Early screening of individuals at high risk of metabolic dysfunction-associated steatotic liver disease (MASLD) facilitates timely intervention and may reduce the risk of progression to cirrhosis. This study aimed to develop a reliable and clinically practical model for MASLD discrimination. METHODS: A total of 179 individuals were recruited, including 80 healthy controls and 99 MASLD patients. Non-targeted and targeted metabolomics were performed to identify and validate candidate biomarkers associated with MASLD. Seven machine learning algorithms were constructed for MASLD discrimination, and diagnostic performance was assessed using accuracy, sensitivity, specificity, and the area under the curve (AUC). Shapley Additive Explanations analysis was used to interpret feature contributions. RESULTS: Non-targeted metabolomics analysis demonstrated that tryptophan metabolism was the most significantly enriched pathway in MASLD. Targeted metabolomics analysis verified that serum tryptophan metabolites were significantly dysregulated between MASLD and healthy controls. According to the ranked feature importance and model performance, a Random Forest-based model combining two tryptophan-related metabolites (kynurenine and neopterin) and five routine clinical tests (triglyceride, high-density lipoprotein cholesterol, uric acid, alanine aminotransferase, and glucose) was established, showing excellent performance with an AUC of 0.926-0.972, a sensitivity of 0.840-0.966, and a specificity of 0.810-0.913. CONCLUSIONS: The developed model integrating targeted tryptophan metabolites with routine clinical tests provides a promising tool to identify individuals at high risk of MASLD. Kynurenine and neopterin emerged as the most informative markers associated with MASLD, providing novel insights for further validation and mechanistic exploration.