Research On Pathogenic Gene Detection Algorithm

With the development of various genetic testing technologies,humans have obtained more and more data,but the number of genes related to specific diseases is very small.The existing ability of humans is still unable to discover the secrets of human pathogenic mechanisms.Research here The content is to make full use of limited resources to unearth genes that may cause disease.Existing algorithms are weak in discovering disease-causing genes of complex diseases,but deep learning can predict them well.The recommendation algorithm and the pathogenic gene found that this article has similarities,so this research has been supplemented and learned.Prediction;Finally,fusion of disease classification information can be integrated to obtain a more effective disease expression,which further improves the predictive ability.1.A disease-causing gene detection algorithm based on feature selection.The problem of pathogenic gene detection is regarded as a feature subset selection problem,and it is abstracted as a mathematical problem,that is,to find the independent variable(gene)that is most related to the dependent variable(gene phenotype).Inspired by the graph message dissemination network,the paper uses the similarity value of the independent variable and the independent variable to be mapped to the weight of the edge in the graph,and the similarity value of the independent variable and the dependent variable is mapped to the value of the node.Way to update.Experimental results show that this method can effectively weaken the similarity between non-pathogenic genes and diseases,and at the same time increase the similarity between pathogenic genes and diseases.2.Information extraction algorithm based on disease gene diversity network.The feature selection scheme can only narrow the range of candidate genes,and cannot accurately predict the relationship of disease genes.Therefore,this paper proposes a disease-causing gene detection algorithm based on graph convolution.The relationship between diseases and genes is formed into a binary heterogeneous network.The difficulty lies in how to use existing algorithms to extract effective vectors from the heterogeneous network and how to train the heterogeneous network.Aiming at these two problems,the spatial map convolution is improved,and a heterogeneous network information extraction model is obtained.The model considers the influence of different types of neighbor nodes during convolution,and uses the wheel rotor graph scheme proposed in this paper for training.The experimental results show that this method can comprehensively use the three-part relationship between disease and gene composition for fusion,and finally obtain a better prediction effect.3.Adaptive disease vector representation.After the above heterogeneous network training,genes and disease vectors can be obtained.The previous inherent deep learning algorithms are to directly calculate the cosine similarity between the two vectors.This calculation is unreasonable,because some diseases are caused by multiple similar genes,and some diseases are caused by multiple dissimilarities.Gene causes,which may cause deviations in calculation results.Therefore,this article proposes a decomposing disease information extraction scheme,which reduces the granularity of the disease and decomposes the granularity of the disease into multiple phenotypes.Focusing on phenotype and gene,the heterogeneous network training method is used to train the phenotype-gene heterogeneous network,and the vectorized representation of each node is obtained.Then the phenotype vector is weighted and summed to obtain the vector representation of the corresponding disease.The specific weight is the similarity measurement value between the phenotype vector and the compared gene vector.Experimental results show that this method can effectively improve the prediction accuracy.