Convolutional Neural Networks for DNA optical mapping

Optical mapping is one of the next generation sequencing methodologies that utilizes DNA sequence specific labeling in order to obtain 'barcodes' which are unique for each species. A novel methodology, termed Fluorocode, has recently been developed that combines basic principles of optical mapping together with microscopy, allowing for high throughput data acquisition and multiple sequence detection by stretching DNA molecules on the surface of a substrate. The acquired images can then be segmented and individual barcodes cross correlated to the reference sequence in the database to obtain a match. While this method has been proven to work well, it is rather slow. Another proposed approach is to use Convolutional Neural Networks (CNN's) for direct segmentation and sequence matching of Fluorocode images, allowing potentially for a substantial increase in the speed of data analysis. The project is aimed at developing the methodology that utilizes CNN's and its comparison to other already developed existing methods.