Analysis of Optical Mapping Data with Neural Network

Abstract

Optical Mapping is a method of DNA sequencing, that can be used to detect large structural variations in genomes. To create these optical maps, a restriction enzyme is mixed with DNA where the enzyme binds to DNA creating labels called restriction sites. These restriction sites and can be captured by fluorescent microscope with a camera. One of the tools that can capture these images is Bionano Genomics Saphyr mapping instrument. Their system produces high-resolution images of DNA molecules with restriction sites and their software detects them. Molecules in these images are visualized by gray lines with restrictions sites appearing brighter. Some of these molecules have very low brightness and with static noise around them, they are almost indistinguishable from the background. This work proposes GPU accelerated method for molecule detection. A neural network was used for molecule segmentation from the image and the dataset for the network was created from the results of Bionano Genomics tools. This detection method can be used as a substitute for restriction map detection in the Bionano Genomics processing pipeline or as a tool that can highlight the regions of interest in the raw optical maps images.