Machine Learning Reveals Serum Glycopatterns as Potential Biomarkers for the Diagnosis of Nonalcoholic Fatty Liver Disease (NAFLD)

Abstract

Nonalcoholic fatty liver disease (NAFLD) has emerged as the predominant chronic liver condition globally, and underdiagnosis is common, particularly in mild cases, attributed to the asymptomatic nature and traditional ultrasonography's limited sensitivity to detect early-stage steatosis. Consequently, patients may experience progressive liver pathology. The objective of this research is to ascertain the efficacy of serum glycan glycopatterns as a potential diagnostic biomarker, with a particular focus on the disease's early stages. We collected a total of 170 serum samples from volunteers with mild-NAFLD (Mild), severe-NAFLD (Severe), and non-NAFLD (None). Examination via lectin microarrays has uncovered pronounced disparities in serum glycopatterns identified by 19 distinct lectins. Following this, we employed four distinct machine learning algorithms to categorize the None, Mild, and Severe groups, drawing on the alterations observed in serum glycopatterns. The gradient boosting decision tree (GBDT) algorithm outperformed other models in diagnostic accuracy within the validation set, achieving an accuracy rate of 95% in differentiating the None group from the Mild group. Our research indicates that employing lectin microarrays to identify alterations in serum glycopatterns, when integrated with advanced machine learning algorithms, could constitute a promising approach for the diagnosis of NAFLD, with a special emphasis on its early detection.