Machine learning-based integration reveals reliable biomarkers and potential mechanisms of NASH progression to fibrosis

Abstract

Abstract Non-alcoholic fatty liver disease (NAFLD) affects about 25% of adults worldwide. Its advanced form, non-alcoholic steatohepatitis (NASH), is a major cause of liver fibrosis, but there are no non-invasive tests for diagnosing or preventing it. In our study, we analyzed data from multiple sources to find crucial genes linked to NASH fibrosis. We built diagnostic models using 103 machine learning algorithms and validated them with two external datasets. All models performed well, with the best one (RF + Enet[alpha = 0.6]) achieving an average AUC of 0.822. This model used five key genes: LUM, COL1A2, THBS2, COL5A2, and NTS. Our findings show that these genes are important in collagen and extracellular matrix pathways, shedding light on how NASH progresses to liver fibrosis. We also found that certain immune cells, like M1 macrophages, are involved in this process. This study provides a reliable diagnostic tool for assessing fibrosis risk in NASH patients and suggests potential for immunotherapy, laying a foundation for future treatments.