Identification of mobile element insertion from whole genome sequencing data using deep neural network model

Mobile element insertions (MEIs) are a major contributor to genome evolution and play an essential role in the regulation of gene expression, as well as being implicated in various human diseases. This study introduces DeepMEI, a tool based on a convolutional neural network model that transforms the MEI identification process into an image recognition problem and automatically learns complex and abstract representations of MEI features in whole genome sequencing data. DeepMEI outperformed existing tools in the benchmark dataset from the Genome in a Bottle consortium, with a precision of 0.90 and recall of 0.70. Moreover, factors such as sequencing depth, ME integrity, and genome mappability can affect MEI identification accuracy. Using DeepMEI, we reanalyzed 3,202 high-coverage whole-genome sequencing samples from the 1000 Genome Project (1kGP) phase 4 release, discovering 1.71-fold more non-reference MEIs, totaling 6,218,088, with 92.2% of the increase coming from rare MEIs (allele frequency <1%). This enhances our understanding of MEIs role in human disease and evolution. The DeepMEI tool and the updated 1kGP MEI dataset can be accessed at https://github.com/xuxif/DeepMEI.