Research On Prediction Method Of Long Non-coding RNA Based On Position Weight Matrix

The research of the Human Genome Project shows that in the human genome,less than 2% of all genome sequences have the function of encoding proteins and the rest are lacking in the ability to encode proteins.which was once regarded as a “junk DNA” gene in the early stage.Until 2004,researchers discovered that the so-called “junk DNA” sequences may hide a large number of DNA regulatory elements,transposons and non-coding RNA genes.After the completion of Encyclopedia of DNA Elements(ENCODE project),it was further discovered that most DNA sequences can be transcribed into RNA,and most of the transcription products are non-coding RNA,while in noncoding RNA,the vast majority of the transcript are long non-coding RNAs longer than 200 nucleotides.In recent years,researchers have continued to heat up the study on these long noncoding RNAs.the results show that RNA can regulate the expression of protein encoding genes at the transcriptional and post-transcriptional level,which is widely involved in important biological processes including cell differentiation and ontogeny,its abnormal expression is closely related with the occurrence of a variety of major human diseases.However,there are still a large number of long non-coding RNAs that are not identified,therefore,how to quickly and accurately select long noncoding RNA from a large number of transcripts is a very worthwhile study subject.In this paper,the method of computer is used to predict and identify long non-coding RNAs.Compared with the biological method,the identification efficiency is greatly improved.There are three major problems in the existing study of long non-coding RNA prediction.First,many prediction methods rely too much on existing species protein-coding library.Once the current species corresponds to a smaller number of protein-coding libraries,the final recognition results will be affected.Second,once there are some sequence errors in the sequencing process,the recognition rate of some prediction methods will be greatly reduced,and sequence errors in the sequencing process are almost inevitable.Third,the location-related information of the sequence could not be collected in the feature extraction,and most of them are the features of nucleotide content or nucleotide combination.In order to solve the above problems,this article first analyzes the characteristics of long non-coding RNA and m RNA sequences,and extracts two major types of features,namely biometric and sequence structure features,in which the biometric class contains open reading frames and polymer features,sequence structure feature class includes k-mer feature,Fickett feature and position weight matrix feature.This feature extraction method does not depend on protein-coding library,and has a certain fault tolerance.In the feature extraction,we used the position weight matrix method for the first time to extract the positional characteristics of nucleotides,and achieved well results in the experiment.In order to improve the training speed of the prediction method and reduce the dimensionality of the feature space,the feature selection method is used in this paper to use packing method and filtering method in order.After feature selection,we obtain a small number of representative features.Next,we use the machine learning algorithms of support vector machine,random forest,and back propagation neural network respectively to train the training set,and use the grid traversal method to select the parameters of the classifier model.After the experiment comparison,we select support vector machine as the classifier of the model for predicting long non-coding RNA.The experimental results show that the prediction model has a better effect.In constructing the long non-coding RNA prediction model,we used a 10-fold cross validation to verify the prediction model on the training set,and finally achieved higher accuracy and consistency on the test set.Compared with the previous long non-coding RNA prediction methods,the prediction results of this paper also have a greater advantage,and this method is not dependent on the protein-coding library of the corresponding species.Experiments were performed on crossspecies datasets.The results show that the proposed algorithm also has certain universality.