Supervised and Unsupervised Classification of lncRNA Subtypes

Many small nucleolar RNAs and many of the hairpin precursors of miRNAs are processed from long non-protein-coding (lncRNA) host genes. In contrast to their highly conserved and heavily structured payload, the host genes feature poorly conserved sequences. Nevertheless there is mounting evidence that the host genes have biological functions. No obvious connections between the function of the host genes and the function of their payloads have been reported. Here we inverstigate whether there is an association of host gene function or mechanisms with the type of payload. To assess this hypothesis we test whether the miRNA host genes (MIRHGs), snoRNA host genes (SNHGs), and other lncRNAs host genes can be distinguished based on sequence and structure features. A positive answer would imply a correlation between host genes and their payload. While the three classes can be distinguished reliably when the classifier is allowed to extract features from the payloads, this is no longer the case when only sequence and structure of parts of the host gene distal from the snoRNAs or miRNA payload is used for classification. We conclude therefore, that MIRHGs and SNHGs do not form separate non-coding RNA classes that are distinct from each other or from lncRNAs without payloads.