Identification of a dinucleotide signature that discriminates coding from non-coding long RNAs

To date, the main criterion by which long ncRNAs (lncRNAs) are discriminated from mRNAs is based on the capacity of the transcripts to encode a protein. However, it becomes important to identify non-ORF-based sequence characteristics that can be used to parse between ncRNAs and mRNAs. In this study, we first established an extremely selective workflow to define a highly refined database of lncRNAs which was used for comparison with mRNAs. Then using this highly selective collection of lncRNAs, we found the CG dinucleotide frequencies were clearly distinct. In addition, we showed that the bias in CG dinucleotide frequency was conserved in human and mouse genomes. We propose that this sequence feature will serve as a useful classifier in transcript classification pipelines. We also suggest that our refined database of "bona fide" lncRNAs will be valuable for the discovery of other sequence characteristics distinct to lncRNAs.