Bioinformatics analysis and machine learning approach applied to the identification of novel key genes involved in non-alcoholic fatty liver disease

Abstract Non-alcoholic fatty liver disease (NAFLD) comprises a range of chronic liver diseases that result from the accumulation of excess triglycerides in the liver, and which, in its early phases, is categorized NAFLD, or hepato-steatosis with pure fatty liver. The mortality rate of non-alcoholic steatohepatitis (NASH) is more than NAFLD; therefore, diagnosing the disease in its early stages may decrease liver damage and increase the survival rate. In the current study, we screened the gene expression data of NAFLD patients and control samples from the public dataset GEO to detect DEGs. Then, the correlation betweenbetween the top selected DEGs and clinical data was evaluated. In the present study, two GEO datasets (GSE48452, GSE126848) were downloaded. The dysregulated expressed genes (DEGs) were identified by machine learning methods (Penalize regression models). Then, the shared DEGs between the two training datasets were validated using validation datasets. ROC-curve analysis was used to identify diagnostic markers. R software analyzed the interactions between DEGs, clinical data, and fatty liver. Ten novel genes, including ABCF1, SART3, APC5, NONO, KAT7, ZPR1, RABGAP1, SLC7A8, SPAG9, and KAT6A were found to have a differential expression between NAFLD and healthy individuals. Based on validation results and ROC analysis, NR4A2 and IGFBP1b were identified as diagnostic markers. These key genes may be predictive markers for the development of fatty liver. It is recommended that these key genes are assessed further as possible predictive markers during the development of fatty liver.