(253bw) Meta-Analysis of Cellular Toxicity of Cadmium-Containing Quantum Dots Using Bayesian Networks
- Conference: AIChE Annual Meeting
- Year: 2016
- Proceeding: 2016 AIChE Annual Meeting
- Group: Computational Molecular Science and Engineering Forum
- Time: Monday, November 14, 2016 - 6:00pm-8:00pm
The rapidly expanding application of engineered nanomaterials (ENMs) across diverse areas, such as medicine, electronics, biomaterials, and energy production, raises concerns about their potential adverse impacts on the environment and health. Accordingly, there is a global drive to ensure that the ENMs developed for various purposes will not pose risk to human health and the environment. In this regards, voluminous data have been generated in high throughput toxicity screening studies to evaluate bacterial and cellular bioactivity induced by exposure to ENMs. Reported toxicity data can be structured or unstructured, quantitative or qualitative, and at different levels of confidence and consistency across different toxicological studies. Currently, efforts to arrive at generalization of ENM toxicity behavior via data-driven models have typically been based on datasets from single studies with little focus on the collective body of published evidence. In general, irrespective of the level of complexity of sought toxicity models, the application of literature data mining and knowledge extraction approaches to heterogeneous datasets must consider the value of information and the body of evidence in order to: (1) evaluate the relevance and significance of various ENMs physicochemical and experimental conditions (i.e., attributes) to toxicity metrics, and (2) develop data-driven models for correlating toxicity metrics with the identified quantitative and qualitative attributes. Accordingly, the present study presents a cause-effect inference approach, suitable for analyzing and extracting knowledge from published body of evidence, focusing on the cellular toxicity of cadmium-containing semiconductor quantum dots (QD). In the present approach Bayesian Networks (BNs) were developed using 2,703 data samples (each with 24 qualitative and quantitative attributes) collected from 448 publications. The BN serve as both a data-driven model for predictive to QD cellular toxicity but can also provide information about the conditional dependence of the toxicity on various ENM properties and exposure conditions. Analyses of the body of evidence via BNs demonstrated that cellular toxicity of QDs closely correlated with their surface properties (including shell, surface charge, ligand and surface modification), diameter, assay type and exposure time. The present work suggests that BNs are effective for integrating qualitative and quantitative data thereby providing the means for interrogating wide-range of toxicity data and for use in predictive toxicology via meta-analysis.