A novel analysis method for biomarker identification based on horizontal relationship: identifying potential biomarkers from large-scale hepatocellular carcinoma metabolomics data.

Abstract

Omics techniques develop quickly and have made a great contribution to disease study. Omics data are usually complex. How to analyze the data and mine important information has been a key part in omics research. To study the nature of disease mechanisms systematically, we propose a new data analysis method to define the network biomarkers based on horizontal comparison (DNB-HC). DNB-HC performs molecule horizontal relationships to characterize the physiological status and differential network analysis to screen the biomarkers. We applied DNB-HC to analyze a large-scale metabolomics data, which contained 550 samples from a nested case-control hepatocellular carcinoma (HCC) study. A network biomarker was defined, and its areas under curves (AUC) in the receiver-operating characteristic (ROC) analysis for HCC discrimination were larger than those defined by six efficient feature selection methods in most cases. The effectiveness was further corroborated by another nested HCC dataset. Besides, the performance of the defined biomarkers was better than that of α-fetoprotein (AFP), a commonly used clinical biomarker for distinguishing HCC from high-risk population of liver cirrhosis in other two independent metabolomics validation sets. All and 90.3% of the AFP false-negative patients with HCC were correctly diagnosed in these two sets, respectively. The experimental results illustrate that DNB-HC can mine more important information reflecting the nature of the research problems by studying the feature horizontal relationship systematically and identifying effective disease biomarkers in clinical practice. Graphical abstract.