Abstract

Background:

Non-alcoholic fatty liver disease (NAFLD) affects a substantial proportion of the general population, but little is known about the impact of specific nutrients on its prevention. Utilizing unbiased machine learning, we aimed to investigate the association between dietary nutrients and the development of NAFLD in the large UK Biobank dataset.

Methods:

We analyzed data from the UK Biobank, focusing on individuals with dietary assessments (up to five dietary questionnaires per person) and excluding those with pre-existing liver disease. Comprehensive benchmarking revealed superiority of a random forest classifier to examine the association between dietary questionnaire nutrients and steatosis development over 11 years of follow-up. A cohort study of more than 200,000 participants was then conducted to assess the association between manganese intake and liver outcomes (ICD10 codes). All analyses were adjusted for age, sex, BMI, Townsend index of socioeconomic status, kcal, alcohol intake, protein intake, fat intake, carbohydrate intake, and multiple testing.

Results:

A random forest classifier was used to analyze the feature importance of 63 nutrients and imaging-proven steatosis in a cohort of over 35,000 UK Biobank participants. Our results showed that participants with higher dietary manganese intake were less likely to have imaging-proven steatosis, suggesting a potential protective effect of manganese against NAFLD (Figure 1 A). We then validated the importance of manganese in a cohort study of over 200,000 UK Biobank participants to examine the relationship between manganese intake and liver outcomes and found that higher manganese intake was associated with a lower risk of ICD-10 coded fatty liver (OR=0.851 (0.804-0.900), p<.001), independent of other potential confounders. In addition, there was a significant effect of manganese on metabolomics available in >49,000 individuals (Figure 1B) and showed an association with reduced VLDL secretion.

Conclusion:

Our study provides evidence that higher manganese intake may be associated with lower odds of steatosis in a large population-based sample. These findings highlight a potential role for manganese in the prevention of NAFLD, but further research is needed to confirm these findings and to investigate the underlying mechanisms. Finally, our study elucidates a comprehensive workflow that leverages the potential of machine learning techniques and vast datasets to facilitate precision nutrition strategies for the prevention of liver disease.

A)Feature Importance for Predicting imaging-diagnosed steatosis using a Random Forest Model: For each input feature, we calculated %IncMSE by permuting the feature's values and measuring increase in the Mean Squared Error.

B)Associations of metabolic biomarkers with manganese intake among >49,000 participants: Hazard ratios (with 95% confidence intervals) presented per 1-SD higher metabolic biomarker.