Research On Genomic Structure Variation Detection Method Based On Deep Learning

In the detection tasks for different types of genomic variation,large-scale structural variation has various characteristics,complex causes,and difficulty in recognition.The results obtained by the existing traditional methods are generally poor in accuracy.In view of the promising prospects of deep learning in the field of high-throughput genomics,this paper introduces deep learning related technologies to study structural mutation detection methods with high accuracy.The main research contents of this article include:(1)A structural variation image coding method combining multiple typical structural variation features is proposed,which defines targeted image drawing forms,stitching rules and scaling methods,which have been proved by experiments to effectively correct the non-intuitiveness of structural variation data and improve subsequent classification Accuracy.(2)A method of genome structure variation detection based on image classification network is proposed.This method uses six different classification network models,including large-scale Image Net-based model and lightweight network model,and conducts network separately according to the characteristics of the model.Details adjustment.By using the weighted training method,the model oscillation can be effectively reduced without affecting the generalization ability,and the network training can be accelerated to reach convergence.Comparative experiments show that this method can achieve a higher accuracy of structural variation detection.(3)A genomic structure variation detection method based on target segmentation network is proposed,which is based on the Two-Stage framework for multi-task detection.By using the intermediate information of the structural variation image coding process,real-time Mask mask annotation is realized.Experiments show that this method can achieve a higher structure type and coverage area determination accuracy.