Lncdml: Identification Of Long Non-Coding Rnas By Deep Metric Learning

The next-generation sequencing technologies provide a great deal of transcripts for bioinformatics research. Specially, because of the regulation of long non-coding RNAs (lncRNAs) in various cellular processes, the research on lncRNAs is in full swing. And the solution of lncRNAs identification is the basis for the in-depth study of its functions. In this study, we present an approach to identify the lncRNAs from large scale transcripts, named lncDML which is completely different from previous identification methods. In our model, we extract signal to noise ratio (SNR) and k-mer from transcripts sequences as features. Firstly, we just use the SNR to cluster the original dataset to three parts. In this process, we achieve preliminary identification effect to some extent. Then abandoning traditional feature selection, we directly measure the relationship between each pair of samples by deep metric learning for each part of data. Finally, a novel classifier based on complex network is applied to achieve the final identification. The experiment results show that lncDML is a very effective method for identifying lncRNAs.