Serum diagnosis of gastric cancer using surface-enhanced desorption ionization mass spectrometry and artificial neural network analyses

Abstract

Objective To develop an artificial neural networks tool and use it to identify proteomic patterns in serum so as to distinguish gastric cancer from controls. Methods Serum samples from 84 gastric cancer patients and 75 controls were randomized into training set (106 samples) and test set (53 samples). At first, samples of the training set were detected using SELDI mass spectrometry and CMIO protein chips. Using a multi-layer ANN with a back propagation algorithm, a proteomic pattern that could distinguish cancer from control samples was identified in the training set. The discovered pattern was then used to determine the accuracy of the classification system in the test set. Results Totally 5 differentially expressed proteins between patients and controls were identified. The five proteins (P < 0.05, m/z at 7567,6742,5262,4869, 4256) were chosen to develop ANN based diagnostic model. The model was blindly tested in the test set for diagnosing gastric cancer. The sensitivity and specificity was 90.0% and 91.3% respectively. Conclusions Combination of SELDI with the artificial neural networks can get a high sensitivity and specificity approach to identify the gastric cancer from the controls. The method shows great potential for early diagnosis of gastric cancer and screening of new tumor biomarkers. Key words: Stomach neoplasms; Spectrometry, mass, matrix-assisted laser desorption-ionization; Neural networks(computer)