Hepatitis B virus (HBV) is the most common DNA virus that may cause hepatitis, cirrhosis and hepatocellular carcinoma. Although many people are persistently infected with HBV, the serum viral load and host immune response varies from person to person. Because the high rate of mutations in HBV protein sequences will alter the protein expressions and even their functions, in this paper, we explore to discover clinical biomarkers of chronic hepatitis B by mining mutation hotspots. A one year follow-up study was conducted with a total number of 1,694 clones from 23 patients with HBeAg-positive chronic hepatitis B. Serum alanine aminotransferase, HBV DNA and HBeAg levels were monthly measured and used as the criteria for clustering the patients into different subgroups. Using monthly derived HBV precore/core protein sequences, we analyzed amino acid mutations responsible for serologic and clinical outcome of each patient subgroup. Using an integration of covariance network and point mutation rule methods, we identified several representative covariance networks of each patient subgroup. Validation with literature-curated mutation hotspots showed that the identified mutations were strongly associated with the viral loads, presence of HBeAg-seroconversion in sera, HBV genotypes and amino acid properties. We further used these identified networks containing mutation hotspots to develop a feature tree, which is applicable for clinicians to prescribe patients a suitable treatment at early stage of HBV infection even though the patients exhibit no obvious symptoms.