Abstract
BACKGROUND: MicroRNAs (miRNAs) have great potential serving as tumor biomarkers and therapeutic targets. As the rapid development of high-throughput experimental technology, gene expression experiments have become more and more specialized and diversified. The complex data structure has brought great challenge for the identification of biomarkers. In the meantime, current statistical and machine learning methods for detecting biomarkers have the problem of low reliability and biased criteria. RESULTS: This study aims to select combinatorial miRNA biomarkers, which have higher sensitivity and specificity than single-gene biomarkers. In order to avoid exhaustive search and redundant information, miRNAs are firstly clustered, then the combinations of representative cluster members are assessed as potential biomarkers. Both the criteria for the partition of clusters and selection of representative members are based on Fisher linear discriminant analysis (FDA). The FDA-based criterion has been demonstrated to be superior to three other criteria in selecting representative members, and also good at refining clusters. In the comparison with eight common feature selection methods, this clustering-based method performs the best with regard to the discriminative ability of selected biomarkers. CONCLUSIONS: Our experimental results demonstrate that the clustering-based method can identify microRNA combinatorial biomarkers with high accuracy and efficiency. Our method and data are available to the public upon request.