Research On MicroRNA Prediction Methods Based On Improved Features And Deep Neural Networks

Non-coding RNAs are a general term for ribonucleic acids that do not encode protein.Although not encode protein,they play an important role in gene expression regulation.Micro RNA is a small non coding RNA with a length of approximately 20 nucleotides,which can participate in the transcriptional regulation and post?transcriptional regulation.It plays an important role in the development and growth of organisms.The abnormal expression of micro RNA,especially the abnormal expression of viral micro RNA,is also closely related to diseases.Therefore,It is of great significance to distinguish the real micro RNA from the pseudo sequence with the similar hair-loop structure.A method for predicting micro RNA is proposed in this paper based on improved features and supervised self-organizing neural network,which attempts to effectively take the advantages of multi-level features and supervised self-organizing mapping.For the query data with given sequence information,we can extract features from the primary sequence and calculate the secondary structure features through the prediction software.All features are fused for subsequent methods.A supervised self-organizing map is a three-layers neural network,with the the input layer for feature extracting,the hidden layer of a self-organizing map,and the output layer computing the two output values of the input sequence labels.The self-organizing map is fully connected to the input layer,learning input data and mapping its high-dimensional spatial distribution information to low-dimensional topological output.The supervised output layer is fully connected to the self organizing map.In the forward transmission,the new features learned by the self organizing map layer are used to calculate the output labels and error values.In the reverse transmission,the error values are returned to the neural network for updating the connection weights.The results show that: 1)Verification experiments are carried out on the date sets of human and viral micro RNA.Which prove that the fusion of structural features and sequence features helps to improve the discriminating power of prediction methods;2)Compare to existing methods on viral micro RNA data,experimental measurement values obtained in the way of crossvalidation illustrate the effectiveness of the method.It is also valid for distinguishing non-coding RNA from coding RNA on different species data sets.The experimental consequent indicates that the method combining multiple features extraction and supervised self-organizing mapping neural network can improve performance of prediction,and have its application prospects.