Gene Variation Detection Approach Based on Multimodal Data Fusion

Structural variations (SVs), related to human’s health, are numerously common in human genes, hence, it’s vital for human to detect the gene structural variation precisely. Traditional gene variation detection methods classify the type and position of gene variation mostly by extracting hand-crafted attributes, which has their limitations. As a well feature extractor, deep convolution neural network could be utilized in gene variation detection field. We present a combined model consisted of Inception-V3 and VGG resembling two-stream model, extracting feature of two different properties at the same time, fusing them and make predictions, achieving 2% improvement averagely in precise, recall and F1-score.