A Deep Learning Model for Automated Quantification of DNA Repair Foci in Somatic Mammalian Cells

Double-strand breaks (DSBs) are the most lethal DNA damages induced by ionising radiations. Here, we present a DSB quantification method based on the analysis of DNA repair foci (pATM and γH2AX) in somatic mammalian cells as the most reliable biomarkers of DSBs. Human dermal fibroblasts and human mesenchymal stem cells were irradiated with 1, 2, and 5 Gy of X-rays and their fluorescent immunocytochemistry microscopy images were captured at 24 h after the irradiation. The captured images were annotated in CVAT platform and the exported data were obtained in YOLO (You Only Look Once) format. The deep learning approach consisted of two stages; a computer vision algorithm and a neural network were used to extract the cells from each image thus providing our features (cell images) and corresponding labels, then these data were split as the required train/test/validation set for the deep learning model based on a YOLO algorithm to count the number of the two mentioned foci per cell. The results of our model were compared with the biologists counting by DARFI software. The developed algorithm and training models were based on the ML/DL/HPC ecosystem of the HybriLIT Heterogeneous Computing Platform.