Non-invasive diagnosis of non-alcoholic steatohepatitis and fibrosis with the use of omics and supervised learning: A proof of concept study

• Published in: Metabolism, clinical and experimental Vol. 101; p. 154005
• Main Authors: Perakakis, Nikolaos, Polyzos, Stergios A., Yazdani, Alireza, Sala-Vila, Aleix, Kountouras, Jannis, Anastasilakis, Athanasios D., Mantzoros, Christos S.
• Format: Journal Article
• Language: English
• Published: United States Elsevier Inc 01.12.2019
• Subjects: Case-Control Studies; Fatty Acids - analysis; Fatty Liver - diagnosis; Female; Glycomics; Humans; Lipidomics; Lipids - analysis; Liver Cirrhosis - diagnosis; Liver fibrosis; Machine learning; Male; Metabolomics; Metabolomics - methods; Metabolomics - trends; Middle Aged; Non-alcoholic fatty liver disease; Non-alcoholic Fatty Liver Disease - diagnosis; Non-alcoholic Fatty Liver Disease - pathology; Non-alcoholic steatohepatitis; Non-invasive; Polysaccharides - analysis; Proof of Concept Study; Supervised Machine Learning; Che; Metabolomics; NAFLD; DG; Glycomics; t-SNE; SVM; Liver fibrosis; Co; AUC; C20:4n6; HDL; LDL; RBF; Ck-18f; C16:0; SM; Non-alcoholic steatohepatitis; NAFL; BMI; AcCa; sPLS-DA; C18:1n9; C18:3n6; Non-invasive; C16:1n7cis; C18:2n6; ROC; HCC; IR; Lipidomics; OvR; kNN; NASH; MCCV; PCA; PA; REF; TG; PC; PE; PG; Machine learning; Non-alcoholic fatty liver disease; SAM; LPC
• DOI: 10.1016/j.metabol.2019.154005
• ISSN: 0026-0495; 1532-8600

Non-alcoholic fatty liver disease (NAFLD) affects 25–30% of the general population and is characterized by the presence of non-alcoholic fatty liver (NAFL) that can progress to non-alcoholic steatohepatitis (NASH), liver fibrosis and cirrhosis leading to hepatocellular carcinoma. To date, liver biopsy is the gold standard for the diagnosis of NASH and for staging liver fibrosis. This study aimed to train models for the non-invasive diagnosis of NASH and liver fibrosis based on measurements of lipids, glycans and biochemical parameters in peripheral blood and with the use of different machine learning methods. We performed a lipidomic, glycomic and free fatty acid analysis in serum samples of 49 healthy subjects and 31 patients with biopsy-proven NAFLD (15 with NAFL and 16 with NASH). The data from the above measurements combined with measurements of 4 hormonal parameters were analyzed with two different platforms and five different machine learning tools. 365 lipids, 61 glycans and 23 fatty acids were identified with mass-spectrometry and liquid chromatography. Robust differences in the concentrations of specific lipid species were observed between healthy, NAFL and NASH subjects. One-vs-Rest (OvR) support vector machine (SVM) models with recursive feature elimination (RFE) including 29 lipids or combining lipids with glycans and/or hormones (20 or 10 variables total) could differentiate with very high accuracy (up to 90%) between the three conditions. In an exploratory analysis, a model consisting of 10 lipid species could robustly discriminate between the presence of liver fibrosis or not (98% accuracy). We propose novel models utilizing lipids, hormones and glycans that can diagnose with high accuracy the presence of NASH, NAFL or healthy status. Additionally, we report a combination of lipids that can diagnose the presence of liver fibrosis. Both models should be further trained prospectively and validated in large independent cohorts. •A lipidomic, glycomic and hormonal analysis was performed in healthy, NAFL and NASH subjects•Results were analyzed with 5 different machine learning techniques in two different platforms•Diagnostic models with excellent accuracy for diagnosing healthy vs. NAFL vs. NASH sumulatneously were developed•A predictive model consisting of 10 lipids showed perfect accuracy for detecting the presence of liver fibrosis