Anti-senescent drug screening by deep learning-based morphology senescence scoring

Dai Kusumoto,Toshiomi Katsuki,Hiromune Sawada,Keiichi Fukuda,Jin Komuro,Akira Kunitomi,Mai Kimura,Tomohisa Seki,Hisayuki Hashimoto,Shogo Ito,Shinsuke Yuasa

doi:10.1038/s41467-020-20213-0

Dai Kusumoto, Toshiomi Katsuki + Show 9 more

Open Access

https://doi.org/10.1038/s41467-020-20213-0

Copy DOI

Abstract

Advances in deep learning technology have enabled complex task solutions. The accuracy of image classification tasks has improved owing to the establishment of convolutional neural networks (CNN). Cellular senescence is a hallmark of ageing and is important for the pathogenesis of ageing-related diseases. Furthermore, it is a potential therapeutic target. Specific molecular markers are used to identify senescent cells. Moreover senescent cells show unique morphology, which can be identified. We develop a successful morphology-based CNN system to identify senescent cells and a quantitative scoring system to evaluate the state of endothelial cells by senescence probability output from pre-trained CNN optimised for the classification of cellular senescence, Deep Learning-Based Senescence Scoring System by Morphology (Deep-SeSMo). Deep-SeSMo correctly evaluates the effects of well-known anti-senescent reagents. We screen for drugs that control cellular senescence using a kinase inhibitor library by Deep-SeSMo-based drug screening and identify four anti-senescent drugs. RNA sequence analysis reveals that these compounds commonly suppress senescent phenotypes through inhibition of the inflammatory response pathway. Thus, morphology-based CNN system can be a powerful tool for anti-senescent drug screening.

Highlights

Advances in deep learning technology have enabled complex task solutions
We previously developed a label-free system to identify endothelial cells among various cell types derived from induced pluripotent stem cells by phasecontrast microscopy images using a CNN6
We established an automated, non-bias quantitative scoring system to evaluate the state of endothelial cells using senescence probability output directly from pre-trained convolutional neural networks (CNN), Deep Learning-Based Senescence Scoring System by Morphology (Deep-SeSMo) (Supplementary Fig. 1a)

Summary

Results

High accuracy identification of senescent cells by a CNN. We induced cellular senescence in human umbilical vein endothelial cells (HUVECs) by using three different stressors: ROS, an anticancer reagent, and replication stress (Supplementary Fig. 1b). The CNN could classify H2O2- or CPT-induced senescent cells and control cells with high accuracy (Supplementary Fig. 3a–d). We used human diploid fibroblasts (HDFs), induced cellular senescence by H2O2 or CPT, cropped input datasets at single-cell resolution levels, and trained the CNN to classify them (Supplementary Fig. 5d). The CNN trained on HUVEC-datasets was able to classify healthy and senescent HDFs (Supplementary Fig. 5g) These results suggest that cellular senescence shows a unique morphologic characteristic, and a morphology-based CNN system can reliably identify senescent cells. A senescence score which was generated by the CNN trained on the datasets acquired at two institutes, Keio and Kyoto (Supplementary Fig. 7b, c), or the CNN trained on another cell type, HDFs (Supplementary Fig. 7d, e), showed high performance.

H-8 ZM 3 36372 TYRPHOSTIN AG 1478

Discussion

Methods

Recall Precision Recall þ Precision ð7Þ

Code availability