BACKGROUND: Although alpha-fetoprotein (AFP) is a useful serologic marker of hepatocellular carcinoma (HCC), it is not sufficiently sensitive to differentiate HCC and liver cirrhosis (LC) caused by hepatitis B virus (HBV) infection. AIMS: The aim is to discover novel noninvasive specific serum biomarkers for the differential diagnosis of HBV-related HCC and LC. METHODS: With a highly optimized peptide extraction and matrix-assisted laser desorption/ionization time of flight/time of flight mass spectrometric approach, we investigated serum peptide profiles of 80 HCC and 67 LC patients. Three supervised machine learning methods were employed to construct classifiers. Receiver operator curves were plotted to evaluate the performance of classifiers. RESULTS: With a support vector machine-based strategy, we picked nine peaks with m/z ratios of 819.49, 1076.14, 1341.72, 2551.44, 3156.44, 3812.88, 4184.26, 4465.92, and 4776.41 to construct the classifier. We proposed a novel method for distinguishing HCC from cirrhosis, based on a multilayer perceptron (MLP) method. We obtained a sensitivity of 90.0%, specificity of 79.4%, and overall accuracy of 85.1% on an independent test set. The combination of the MLP model and serum AFP level outperformed serum AFP marker alone in distinguishing HCC patients from LC patients. In this experience, sensitivity increased from 62.5% to 87.5%, and specificity increased from 79.4% to 88.2%. CONCLUSIONS: Our results indicate that the MLP model is a novel and useful serum peptide pattern for distinguishing HCC and LC. The peptidome signature alone or together with serum AFP determination may be a more effective method for early diagnosis of HCC in patients with HBV-related LC.
BACKGROUND: Although alpha-fetoprotein (AFP) is a useful serologic marker of hepatocellular carcinoma (HCC), it is not sufficiently sensitive to differentiate HCC and liver cirrhosis (LC) caused by hepatitis B virus (HBV) infection. AIMS: The aim is to discover novel noninvasive specific serum biomarkers for the differential diagnosis of HBV-related HCC and LC. METHODS: With a highly optimized peptide extraction and matrix-assisted laser desorption/ionization time of flight/time of flight mass spectrometric approach, we investigated serum peptide profiles of 80 HCC and 67 LC patients. Three supervised machine learning methods were employed to construct classifiers. Receiver operator curves were plotted to evaluate the performance of classifiers. RESULTS: With a support vector machine-based strategy, we picked nine peaks with m/z ratios of 819.49, 1076.14, 1341.72, 2551.44, 3156.44, 3812.88, 4184.26, 4465.92, and 4776.41 to construct the classifier. We proposed a novel method for distinguishing HCC from cirrhosis, based on a multilayer perceptron (MLP) method. We obtained a sensitivity of 90.0%, specificity of 79.4%, and overall accuracy of 85.1% on an independent test set. The combination of the MLP model and serum AFP level outperformed serum AFP marker alone in distinguishing HCC patients from LC patients. In this experience, sensitivity increased from 62.5% to 87.5%, and specificity increased from 79.4% to 88.2%. CONCLUSIONS: Our results indicate that the MLP model is a novel and useful serum peptide pattern for distinguishing HCC and LC. The peptidome signature alone or together with serum AFP determination may be a more effective method for early diagnosis of HCC in patients with HBV-related LC.