Machine learning identifies microbiome associations with MASLD biomarkers

Background: Gut microbiome plays a crucial role in liver health through gut-liver axis. It may contribute to metabolic-dysfunction-associated steatotic liver disease (MASLD), a global concern in individuals without significant alcohol consumption. MASLD is driven by metabolic factors like obesity and dyslipidemia. Investigating the link between the microbiome and MASLD-biomarkers (BMI, triglycerides) could help in detecting and preventing MASLD via microbiome modulation.

Methods: Lifelines biobank's data, including over 16,700 participants, is utilized to extract 422 individuals with 16S rRNA gene amplicon sequencing and phenotypic data (age, gender, BMI, triglycerides, dietary intake). Due to the lack of data on the presence of MASLD ICD codes, BMI≥25 kg/m² and triglycerides≥1.7 mmol/L are used to identify patients at risk of MASLD. We then used machine learning models like Random Forest and XGBoost, with microbiome features (Shannon index, relative phyla abundance) and phenotypic data to define predictors of obesity and hypertriglyceridemia.

Results: Firmicutes are found to be more abundant in obese patients, while Bacteroidetes are less abundant. For obesity, Random forest performed better with an AUC of 0.72 on test set (83 samples). For hypertriglyceridemia, XGBoost outperformed Random Forest with an AUC of 0.74 on test set (83 samples). Shannon index and microbiome phyla, including Bacteroidetes, Actinobacteria, and Desulfobacterota, are identified to be among top 7 features in classifying BMI and triglycerides and might also be important for MASLD.

Conclusion: The study highlights the potential of identifying microbiome-based MASLD-biomarkers, and the role of machine-learning. Still, biopsy validation in MASLD is needed.

Publication History

Article published online:
20 January 2025