A Serum Protein Profiling Study On Primary Hepatic Carcinoma With Artificial Neural Network Diagnosis Model

Objective:primary hepatic carcinoma (PHC) is the cancer of hepatocyte or Intrahepatic bile duct, which is frequent malignant tumor and has the third death rate in alimentary system tumor, and second only to gastric cancer and esophageal cancer. Hepatocellular carcinoma (HCC) is the most frequent occurrence in PHC. And total five survival rate is lower than 5% in the world wide. Because of lacking of early diagnosis methods, the case fatality rate becomes more and more highly in China. Many patients loss their best opportunity to cure. We need exploring a new quick, simple, high sensitivity, good specificity early diagnosis method. Surface enhanced laser desorption ionization time of flight mass spectrometry (SELDI-TOF-MS) is the most popular technology with high sensitivity and specificity in those test facilities. When any disease showes the pathological change, the composition and quantity of proteins will change in cells, and pattern of proteins can reflect the changing. So we could screen these relative proteins of tumor with comparing on different patients protein in serum. In this study, using SELDI-TOF-MS to detect hepatitis B, liver cirrhosis, primary liver cancer, other gastrointestinal tumors and health human serum protein profiling with proteomics approach, and screening in patients with primary liver cancer-specific expression of serum protein markers, combined with artificial neural network (ANN) to establish prediction model for primary liver cancer and explored the clinical value of laboratory diagnosis in order to establish an effective early detection of hepatocellular carcinoma objective experimental index. Methods:The 435 serum samples were tested by SELDI-TOF-MS matched with Gold Chip and finded different proteins through Biomarker Wizard software. Using ANN to develop a model of diagnosis of PHC.75 patients with type B hepatitis,68 patients with hepatic cirrhosis,100 patients with PHC,91 patients with other alimentary system tumor and 101 health controls were included. Samples were randomly assigned into two subsets, the training set (35 patients with type B hepatitis,33 patients with hepatic cirrhosis,50 patients with PHC,41 patients with alimentary system and 51 health controls), and the testing set (40 patients with type B hepatitis,35 patients with hepatic cirrhosis,50 patients with PHC, 50 patients with alimentary system and 50 health controls). The serum samples were binded to the gold chip and detected serum protein profiling data, and corrected and analysed protein profiling data by Ciphergen ProteinChip 3.0 software. The training set was used for identifying the statistically significant peaks as well as for developing ANN model. And the testing set was used for blind test to validate the diagnostic efficiency of ANN model. Results:Serum protein fingerprint selected 75 differentially expressed proteins than the peak load (P<0.05) in PHC and control groups. The use of which 7 are significant differences in the expression of marker proteins (P<0.01) to establish artificial neural networks diagnosis model, the mass charge ratio (m/z) were 4207,6604,7734,8106,8545,8599,8894. And they correspond to Peptide YY-like, 50S ribosomal protein L30,50S ribosomal protein L35, Neutrophil-activating peptide 2(74), Acyl carrier protein,30S ribosomal protein S21, UPF0330 protein TK1752 respectively by Swiss-Prot roughly. Using blind method to predict the model of PHC, we could get the sensitivity and specificity were 84.00% and 81.25% respectively, and the area under (AUC) of receiver operating characteristic curve (ROC curve) was 0.847, negative predictive value(NPV) was 94.20%, positive predictive value (PPV) was 58.33%, accuracy(ACC) was 81.90%. Conclusions:There were significantly different proteins in PHC, and the ANN model provides a new method of differential diagnosis of PHC. The method is important for differential diagnosis and treating PHC.