Innovative approaches to metabolic dysfunction-associated steatohepatitis diagnosis and stratification

Abstract

The global rise in Metabolic dysfunction-associated steatotic liver disease (MASLD)/Metabolic dysfunction-associated steatohepatitis (MASH) highlights the urgent necessity for noninvasive biomarkers to detect these conditions early. To address this, we endeavored to construct a diagnostic model for MASLD/MASH using a combination of bioinformatics, molecular/biochemical data, and machine learning techniques. Initially, bioinformatics analysis was employed to identify RNA molecules associated with MASLD/MASH pathogenesis and enriched in ferroptosis and exophagy. This analysis unveiled specific networks related to ferroptosis (GPX4, LPCAT3, ACSL4, miR-4266, and LINC00442) and exophagy (TSG101, HGS, SNF8, miR-4498, miR-5189–5p, and CTBP1-AS2). Subsequently, serum samples from 400 participants (151 healthy, 150 MASH, and 99 MASLD) underwent biochemical and molecular analysis, revealing significant dyslipidemia, impaired liver function, and disrupted glycemic indicators in MASLD/MASH patients compared to healthy controls. Molecular analysis indicated increased expression of LPCAT3, ACSL4, TSG101, HGS, and SNF8, alongside decreased GPX4 levels in MASH and MASLD patients compared to controls. The expression of epigenetic regulators from both networks (miR-4498, miR-5189–5p, miR-4266, LINC00442, and CTBP1-AS2) significantly differed among the studied groups. Finally, supervised machine learning models, including Neural Networks and Random Forest, were applied to molecular signatures and clinical/biochemical data. The Random Forest model exhibited superior performance, and molecular features effectively distinguished between the three studied groups. Clinical features, particularly BMI, consistently served as discriminatory factors, while biochemical features exhibited varying discriminant behavior across MASH, MASLD, and control groups. Our study underscores the significant potential of integrating diverse data types to enable early detection of MASLD/MASH, offering a promising approach for non-invasive diagnostic strategies.

Keywords

• MASLD
• MASH
• Ferroptosis
• Exophagy
• Macroautophagy
• Machine learning