IMDSVs: an Integrated Method Based on Machine Learning and Deep Learning of Calling Structural Variations from Long-Read Data

Calling structure variations (SVs) in genome sequencing data is a hot topic in bioinformatics. In this paper, we proposed a new method, called IMDSVs, to solve this problem. IMDSVs combines machine learning method and deep learning method. It uses machine learning method to preprocess the genome file and then maps the processed genome file into image data as the input of the deep learning method. Finally, using image data to train the deep learning method to get the final calling results. We evaluate the results of IMDSVs and other state-of-art calling tools on simulate and real data. The results show that IMDSVs have higher recall and F1-score on different coverage depths, different variant lengths, and different genotypes.