Genomic structure variation prediction method based on convolution and recurrent network

Predict genomic variants from gene sequencing data is the central task in biological genome sequence analysis. It also serves as the foundation for identifying and screening pathogenic variants as well as conducting pharmaco genomics research. The data in the field of genomics is typically massive, high-dimensional, and serialized, and deep learning, as a data- driven algorithm, has strong feasibility and potential in the field of bioinformatics. Based on previous research, the goal of this study is to predict structural variation in high-throughput sequencing data from the 1000 Genome Project’s BAM file NA12878. BAM files are also combined with VCF files to improve prediction efficiency. VCF files are frequently used to store prediction results. It contains information such as the sample number, chromosome position, mutation type, and mutation breakpoint. BAM and VCF files are converted into images in this paper, and a gene structure mutation prediction method based on the fusion of the Inception-ResNet-v2 and BiLSTM algorithm models is proposed.