Inferring of miRNAs as biomarker via subspace dimensionality reduction and clustering

microRNAs (miRNAs) play a important role in a wide range of biological processes by regulating the expression of target genes, and the expression of miRNAs vary significantly among different different tissues and timing of biological process. The specific expression of miRNA creates signatures for a variety of diseases which is important for disease diagnosis and treatment. In this study, a novel hybrid approach that combines dimensionality reduction and clustering analysis is developed to infer the miRNAs as potential biomarkers of corresponding disease. Specifically, Locally Linear Embedding method and density-based clustering method are applied to eight miRNA expression profiles iteratively to acquire informative subspaces and select high-frequency miRNAs in subspaces. The coexpressed miRNAs are detected in each subspace, and the clustering frequency of each miRNA is counted to determine whether the miRNA could be taken as the biomarker for specific disease. As a result, the miRNA co-expression network was constructed to inferr biomarkers, and the differential expression of these biomarkers have been detected under different samples. Furthermore, most of the detected biomarkers have been validated by published literatures, demonstrating the ability of proposed method for inferring biomarker miRNAs.