Using Chemical-Induced Gene Expression in Cultured Human Cells to Predict Chemical Toxicity

Abstract

Chemical toxicity is conventionally evaluated in animal models. However, animal models are resource intensive; moreover, they face ethical and scientific challenges because the outcomes obtained by animal testing may not correlate with human responses. To develop an alternative method for assessing chemical toxicity, we investigated the feasibility of using chemical-induced genome-wide expression changes in cultured human cells to predict the potential of a chemical to cause specific organ injuries in humans. We first created signatures of chemical-induced gene expression in a vertebral-cancer of the prostate cell line for ∼15,000 chemicals tested in the US National Institutes of Health Library of Integrated Network-Based Cellular Signatures program. We then used the signatures to create naı̈ve Bayesian prediction models for chemical-induced human liver cholestasis, interstitial nephritis, and long QT syndrome. Detailed cross-validation analyses indicated that the models were robust with respect to false positives and false negatives in the samples we used to train the models and could predict the likelihood that chemicals would cause specific organ injuries. In addition, we performed a literature search for drugs and dietary supplements, not formally categorized as causing organ injuries in humans but predicted by our models to be most likely to do so. We found a high percentage of these compounds associated with case reports of relevant organ injuries, lending support to the idea that in vitro cell-based experiments can be used to predict the toxic potential of chemicals. We believe that this approach, combined with a robust technique to model human exposure to chemicals, may serve as a promising alternative to animal-based chemical toxicity assessment.

An Overview of Machine Learning and Big Data for Drug Toxicity Evaluation

“…Gene expression data from the LINCS database has been integrated with chemical organ toxicity data in humans from SIDER and OFFSIDES databases to predict organ toxicity. Using a modified naive Bayes model, AUC metrics were evaluated and reported for liver (0.82), heart (0.77), and kidney toxicity (0.76) . Other databases also exist that curate published literature, existing drug studies, and other web resources to infer compound-related gene expression and interaction data such as the Comparative Toxicogenomics Database (CTD) and Drug Gene Interaction Database (DGIdb). , …”

Machine Learning Based Toxicity Prediction: From Chemical Structural Description to Transcriptome Analysis

“…One can induce changes of whole genome expression of cultured human cells of a specific type by adding a test drug to the culture. By analyzing changes in the transcriptome, toxicity of the drug to the cell type, and to the corresponding organ, can be predicted [131]. Schwartz et al used both toxic and non-toxic compounds to treat 3D-cultured human pluripotent stem cell-derived neural cells, then used RNA-Seq to determine the whole genome expression profile, and then used SVM to classify the chemicals according to their toxicity.…”

Prioritization of Contaminants of Emerging Concern in Wastewater Treatment Plant Discharges Using Chemical:Gene Interactions in Caged Fish

“…A growing list of xenobiotics including legacy contaminants (such as metals and persistent organic pollutants) and chemicals of emerging concern (CEC) such as pharmaceuticals, personal care products, flame retardants, industrial chemicals, and pesticides are being released into the environment. , Because the potential environmental hazards of a large number of CECs in surface waters have not been well characterized, a number of screening and prioritization approaches have been developed to assess their potential impacts. − Approaches for assessing CEC impacts on fish health have included determining chemical concentrations in water relative to the amount established as causing a biological effect or Exposure Activity Relationships (EAR), , in vitro assays, ,,, laboratory exposures of fish to field-collected water samples and direct exposure of fish to surface water via cages or flow through systems. ,− Gene expression changes have been employed to assess effects of complex chemical mixtures on fish, such as the potential of wastewater treatment plant (WWTP) discharge to cause estrogenic effects, identify differences in exposure sites, ,,,,, and infer the potential of specific chemicals to cause effects through the use of prior knowledge of chemical:gene or chemical:protein interactions. ,, …”

“…3,12−19 Gene expression changes have been employed to assess effects of complex chemical mixtures on fish, such as the potential of wastewater treatment plant (WWTP) discharge to cause estrogenic effects, 20 identify differences in exposure sites, 12,13,15,16,19,21 and infer the potential of specific chemicals to cause effects through the use of prior knowledge of chemical:gene or chemical:protein interactions. 19,22,23 A complicating factor in applying existing knowledge to infer impacts of specific chemicals in complex mixtures is the limited availability of chemical:gene interaction data for many chemicals. Existing databases of chemical:gene interactions, such as the Comparative Toxicogenomics Database (CTD), 24 STITCH (Search Tool for Interacting Chemicals), 25 or Ingenuity Knowledge Base, 26 are biased toward well studied chemicals and relevant interactions are under-represented for less studied compounds.…”