Prediction of chronic hepatitis B, liver cirrhosis and hepatocellular carcinoma by SELDI-based serum decision tree classification

Publisher Website
Google Scholar
Abstract
To screen potential serological biomarkers and develop decision tree classifications of chronic hepatitis B, liver cirrhosis (LC) and hepatocellular carcinoma (HCC), respectively, with high prediction score for improving diagnosis of liver diseases. The total serum samples were randomly divided into three training sets (41 HBV and 35 health; 36 LC and 35 health; 39 HCC and 35 health) and three testing groups (34 HBV and 38 health; 18 LC and 52 health; 42 HCC and 47 health). Selected WCX2 protein chip capture followed by SELDI-TOF-MS analysis was applied to generate the serum protein profiles. Subsequently serum protein spectra were normalized and aligned by Ciphergen SELDI Software 3.1.1 with Biomarker Wizard including baseline subtraction, mass accuracy calibration, automatic peak detection. Once the intensities of selected significant peaks from the training data set were transferred to further BPS analysis, an optimized classification tree with sequence-decision was established to divide training data set into disease group and control group successfully. A double blind test was employed to determine the clinical sensitivity and clinical specificity of three models. After comparative analysis of SELDI based serum protein profile between the cases of disease and healthy, a HCC decision tree classification with sensitivity of 94.872% and specificity of 94.286%; a LC decision tree classification with sensitivity of 91.667% and specificity of 94.286% and a HBV decision tree classification with sensitivity of 95.122% and specificity of 94.286% were produced by BPS respectively. When three decision tree models were challenged by the double-blind test samples, clinical sensitivity and clinical specificity of these models were predicted in diagnosis of three liver diseases (HCC: 90.48 and 89.36%; cirrhosis: 100 and 86.5%; HBV: 85.29 and 84.21%). SELDI-based decision tree classifications showed great advantages over conventional serological biomarkers in the diagnosis of chronic hepatitis B, LC as well as HCC.