Research On SNP Function Of Non-coding Region Based On Deep Learning

By conducting Genome-wide association study(GWAS),people can locate single nucleotide polymorphisms(SNPs)within the entire human genome and screen out SNPs related to diseases.SNP is a type of genetic variation that occurs when a single nucleotide in the DNA sequence is altered.In GWAS studies,the vast majority of SNP sites are found in non-coding regions.Unlike coding regions that can be transcribed and translated into proteins to perform corresponding functions,genes in non-coding regions affect the life activities of organisms indirectly through regulation.Therefore,the study of the function of non-coding region SNPs has become an important and challenging problem in the field of genomics.This article focuses on predicting nearly a thousand functional features of SNPs in non-coding region of DNA sequence.These features are binary variables with a value of 0 or 1.With the increase of genomics data and the development of deep learning,many studies in the field of genomics have begun to use deep learning methods.This article constructs a deep neural network framework for the prediction of SNP function in non-coding region.The algorithmic framework is called Deep At GRU.Deep At GRU combines convolutional neural network and recurrent neural network,innovatively uses bidirectional GRU with self-attention to extract remote information in DNA sequences.It achieves better prediction results in multi-task classification problems for non-coding SNP functionality.Based on one million training data,8,000 validation data and 10,000 testing data,this article trains Deep At GRU and obtains evaluation metrics that are superior to the three existing algorithmic frameworks called Deep SEA,Dan Q,and Deep GRU.Then,this article uses the Deep LIFT method to extract the key DNA motifs identified by Deep At GRU during the training process.These DNA motifs have a significant correlation with non-coding SNP functional features,such as transcription factor binding sites,providing a genomic explanation for the effectiveness of Deep At GRU.