Prediction And Analysis Of Cancer Differential Genes Based On Alternativate Splicing

Alternative splicing is an important mechanism of protein diversity in a wide range of organisms,which plays an important role in the fine regulation of cell proliferation,differentiation,development,apoptosis and a series of important biological processes.In recent years,it has been found that alternative splicing is closely related to the production of complex diseases including tumors,and the occurrence of some diseases is often accompanied by the disordered expression of splicing isoforms.The existing analysis of breast cancer subtypes is mainly based on single splicing isoform,and the difference in the overall distribution of splicing isoforms caused by alternative splicing disorders is not considered.Consider the shortcomings of existing methods,first we propose a differential analysis method of alternative splicing based on the median value by Jensen-Shannon divergence.The basic idea is to construct the representative expression vector of gene isoforms based on the expression median of cancer and normal samples,and then calculate the JS divergence of the two states according to the percentage of each isoform in the representative vector.Next,we propose a method to predict the subtypes of breast cancer based on JS divergence,which mainly used JS divergence to find genes with large differences in alternative splicing disorders between subtypes,then constructed BP neural network model to classify breast cancer subtypes.The results show our methods can find plenty of genes with significant differences in the overall distribution of splicing isoforms.These genes are not only concentrated in some cancer related pathways,but also in some signaling pathways based on alternative splicing regulation,cell division process and protein function.Compared with the gene-level differential analysis,the genes with significant difference in alternative splicing have better performance in survival analysis.In addition,our methods also have good identification results in the classification of breast cancer subtypes,with a F1-score of 0.89,and can provide personalized drug recommendations for patients with breast cancer subtypes.In conclusion,our methods solve the calculation problem of unpaired samples by constructing the representative expression vector of gene isoforms,can find unique genes in the overall distribution of cancer splicing isoforms,and lay a foundation for further revealing the mechanism of alternative splicing in cancer.