Comet assay annotation data for comet segmentation and scoring (1037 comet assay images)

Comet assay annotation images used in the following paper

Hong, Y., Han, HJ., Lee, H. et al. Deep learning method for comet segmentation and comet assay image analysis. Sci Rep 10, 18915 (2020). https://doi.org/10.1038/s41598-020-75592-7

DeepComet online service website: https://ad3.io/

sign up->dashboard->APPs->Deep Comets->new job

Two biology researchers who were experienced in the comet assay for over half a year under the guidance of a toxicology expert in the assay annotated the images independently and compared each other’s work to ensure consistency. Here, the VGG Image Annotator sofware was used to annotate the images manually. We annotated the intact comet heads without a tail using a circular region tool that outputs the center and radius of the circle. Further, comets with a tail and DNA damage were annotated with a polyline tool, where the frst dot was marked at the center of a comet head, and the second dot was marked at the bottom of the comet head. Tese two points were used to identify the head of a comet. Subsequently, dots were marked along the boundary of the comet tail counterclockwise from the second dot. Te total number of dots were at least eight per each comet.

Comets can be sorted into non-ghost cells with a distinct head and tail, and ghost cells with a small or no nucleoid head and a broad tail. The ghost cells are usually classifed as the highest degree of DNA damage by visual scoring. Until now, there is a controversy over the cause of ghost cells and how they should be analyzed appropriately. Therefore, our method is designed to classify these two kinds of cells, so one can use some custom analysis techniques for ghost cells.

The comets in our datasets were all classifed as non-ghost or ghost cells. Furthermore, we assigned tags to the comets if they were overlapped or located at the boundary of an image (i.e., only a part of the comet is visible). After manual annotation with dots and tags (in the red box), each dotted contour was flled up to give mask images.