Few non-invasive models were established for precisely identifying the immune tolerant (IT) phase from chronic hepatitis B (CHB). This study aimed to develop a novel approach that combined next-generation sequencing (NGS) and machine learning algorithms using our recently published viral quasispecies (QS) analysis package. 290 HBeAg positive patients from whom liver biopsies were taken were enrolled and divided into a training group (n = 148) and a validation group (n = 142). HBV DNA was extracted and QS sequences were obtained by NGS. Hierarchical clustering analysis (HCA) and principal component analysis (PCA) based on viral operational taxonomic units (OTUs) were performed to explore the correlations among QS and clinical phenotypes. Three machine learning algorithms, including K-nearest neighbour, support vector machine, and random forest algorithm, were used to construct diagnostic models for IT phase classification. Based on histopathology, 90 IT patients and 200 CHB patients were diagnosed. HBsAg titres for IT patients were higher than those of CHB patients (p < 0.001). HCA and PCA analysis grouped IT and CHB patients into two distinct clusters. The relative abundance of viral OTUs differed mainly within the BCP/precore/core region and was significantly correlated with liver inflammation and fibrosis. For the IT phase classification, all machine-learning models showed higher AUC values compared to models based on HBsAg, APRI, and FIB-4. The relative abundance of viral OTUs reflects the severity of liver inflammation and fibrosis. The novel QS quantitative analysis approach could be used to diagnose IT patients more precisely and reduce the need for liver biopsy.
Keywords: Chronic hepatitis B; clinical pathology; decision support techniques; machine learning; natural history; quasispecies.