S5: High-throughput methods for accurate prediction

High-throughput methods for the accurate prediction of human organ-specific toxicities

High-throughput methods for accurate prediction, OpenTox Asia 2017

Daniele Zink 


 Institute of Bioengineering and Nanotechnology (IBN)


Principal Research Scientist 


Sijing Xiong1, Faezah Hussain1, Ah Wah Lam2, Jacqueline Chuah1, Peng Huang1, Yao Li1, Lit-Hsin Loo2,3 and Daniele Zink1

1Institute of Bioengineering and Nanotechnology, 31 Biopolis Way, The Nanos, Singapore 138669, Singapore

2Bioinformatics Institute, 30 Biopolis Street, #07-01 Matrix, Singapore 138671, Singapore

3 Department of Pharmacology, Yong Loo Lin School of Medicine, National University of Singapore, 10 Medical Drive, Singapore 117597, Singapore



Evaluating the toxicity of drug candidates, environmental toxicants, industrial chemicals, food additives, cosmetics ingredients, natural compounds and nanomaterials requires predictive methods. Animal models are of limited predictivity and are too slow and costly for screening of the large and increasing numbers of compounds that need to be tested. Also, changes in legislation (e.g. animal bans for cosmetics testing) and other developments steeply increase the demand for alternative methods. However, many alternative methods are of unknown predictivity, and accepted alternative methods for predicting toxicity for human internal organs are not available. This problem is addressed by our work, which was initially focused on the kidney. Recently, we have developed the first predictive animal-free renal platforms, which include predictive models based on human induced pluripotent stem cells and a predictive high-throughput platform (Li et al., 2013; Li et al., 2014; Su et al., 2014; Kandasamy et al., 2015; Su et al., 2015).

The high-throughput platform combines high-content imaging of human renal cells with image-based phenotypic profiling and machine learning. This platform is currently applied in collaboration with the US Environmental Protection Agency to predict the human nephrotoxicity of ToxCast compounds. The test balanced accuracy of the high-throughput platform ranges between ~ 80% - 90% depending on the cell type used. Its predictivity does not depend on the chemical structure and nature of the tested compounds. This platform also gives unbiased and unexpected insights into injury mechanisms and compound-induced cellular pathways. Based on a similar methodology we are now developing high-throughput platforms for predicting toxicity for other human organ systems, including liver and vasculature. Our goal is to develop a portfolio of regulatory accepted predictive alternative methods that cover major human organ systems.