Classifying MicroRNAs by Gabor Filter Features from 2D Structure Bitmap Images on a Case Study of Human MicroRNAs

MicroRNAs are recently discovered small, 20–24 basepairs long non-coding RNA molecules in plants and animals. They regulate many genes at nanoscale by hybridizing to other RNA or DNA molecules. MicroRNAs are derived from longer, 150–500 basepairs or longer hairpin-like precursor RNA structures resulting from folding and pairing of the molecule on itself. Limited sequence similarity hinders discovery and clustering of microRNA precursors from genomes. We introduce a novel method of classifying 2D shapes of simulated, thermodynamically optimal folded RNA structures by bitmap image analysis methods, exemplified here by Gabor filter feature extraction. This approach represents a simplified hypothesis-less ab-initio discovery tool for relevant novel structural features from 2D representations of RNA molecular structures using various image analysis algorithms. It bypasses the need for complex 3D structural data comparisons similar to protein threading methods (e.g., sequence similarity guided 3D folding to a known structure). Results with an example of 222 human microRNA precursors are promising, suggesting the method can reveal new and useful discriminatory features of RNA molecules. This can be useful additional information in multimodal data integration for improved microRNA gene discovery and classification. Prospects for expanding the methodology for other types of molecules and utilization of other image analysis methods are discussed, as well as possibilities for exhaustive genome-wide surveys for discovery and classification of novel non-coding RNA molecules from various organisms.