Research On Computational Prediction Of Deafness Genes

Sudden deafness is a common and frequently occuring disease in otolaryngology.Many studies have shown that sudden deafness is closely related to genetic deafness gene mutations.At present,biological methods are mainly used to identify deafness genes,but traditional biological experiments were costly,time-consuming,and labor-consuming,which made patients easily miss the best time of treatment in medical diagnosis.How to quickly and accurately identify deafness genes and their deaf-causing mutation regions is of great significance for the diagnosis and treatment of deafness.In recent years,computational methods have been widely used in the prediction of cancer,and other pathogenic genes,but there is no computational method in the prediction of deafness genes.Therefore,in order to realize the early diagnosis of deafness,this paper took deafness genes as the analysis object,extracted sequence-based features,and constructed models through machine learning algorithm to predict new deafness gene and its deafness mutation region.In this paper,the support vector machine(SVM)and the back propagation neural network(BPNN)were used.However,a single classification algorithm has some defects.So as to improve the classification performance,a hybrid classifier(BPNN-SVM)was proposed by integrating the two single classification algorithms.To verify the classification performance of the hybrid classifier,two kinds of experiments were designed and completed in this paper:Used the hybrid classifier to predict deafness genes.In this experiment,the 53-dimensional features based on deafness gene sequences and protein sequences were first calculated;then the feature set was divided into training set,validation set and test set according to the ratio of 6:2:2,and they were used to construct a BPNN-SVM hybrid classifier model,and finally evaluated the performance of the model based on the classification results.The results were as follows: in the first test bench data set,6 genes were predicted to be deafness genes out of 62 known genes;in the second test bench data set,3 known genes were predicted to be deafness genes;in the third test bench data set,2genes were confirmed to be deafness genes out of 18331 unknown genes by the model.The results showed that the BPNN-SVM hybrid classifier model based on sequence features was valuable for identifying novel deafness genes from unknown genes.Used the hybrid classifier to predict the deafness gene mutation regions.The purpose of experiment one was to predict novel deafness genes from unknown genes,but the deaf gene diagnosis needs to identify the deafness mutation site of deafness genes.Limited to the current data resources and analysis models,we can not use the models to identify mutation sites.The experiment took gene coding sequences as the analysis object,calculated the codon bias characteristics,Hurst exponent,and newly introduced the features based on information theory,and obtained 41 dimension features.The process and evaluation method of model construction were similar to those of experiment one.In the experiment,we constructed a set of data to be predicted to verify the performance of the BPNN-SVM hybrid classifier model.The results showed that one of the predicted top20 mutation regions was confirmed to contain deafness mutation sites.It was indicated that the BPNN-SVM hybrid classifier based on sequence features could provide meaningful auxiliary information for the recognition of deafness gene mutation regions,and had the potential to be further applied in practice.