Abstract 13576: Anti-senescent Drug Screening by Deep Learning-based Morphology Senescence Scoring

Abstract

Backgrounds: Deep learning is currently being applied to many complex tasks. The accuracy of image classification, in particular, has improved with the development of convolutional neural networks (CNNs). Cellular senescence is a hallmark of ageing and is highly important for the pathogenesis of aging-related diseases. Furthermore, cellular senescence has been identified as a potential therapeutic target. Specific molecular markers are widely used to identify senescent cells, but senescent cells also show unique morphology, which can be identified by CNN. Objective: We develop a strong-performing, morphology-based CNN system called Deep-SeSMo to identify senescent cells, and establish a novel quantitative scoring system to evaluate the probability that an endothelial cell is senescent. Methods and results: We induced cellular senescence in human umbilical vein endothelial cells (HUVECs) using three different stressors: ROS; an anti-cancer reagent; and replication stress, and demonstrated that CNN performed with high accuracy and generalizability. Then, we developed a “senescence score” based on the output of the pre-trained CNN, optimized it for the classification problem, and defined the overall average output probability calculated by the pre-trained CNN as the quantitative senescence score, namely Deep-SeSMo. Deep-SeSMo correctly evaluates the effects of well-known anti-senescent reagents; nicotinamide mononucleotide (NMN) and metrofmin. We screened for drugs that control cellular senescence using a kinase inhibitor library by Deep-SeSMo-based drug-screening and discovered four novel anti-senescent drugs. RNA sequence analysis revealed that those compounds commonly suppress senescent phenotypes through inhibition of the inflammatory response pathway. Conclusions: Using Deep-SeSMo, we demonstrated that the morphology-based CNN system can be a powerful tool for anti-senescent drug screening.