This article presents a dose-response clustering algorithm for the purpose of revealing toxicologically distinct clusters of carbon nanotubes (CNTs). Current exposure guidelines such as those published by the National Institute for Occupational Safety and Health (NIOSH) consider all types of CNTs and carbon nanofibers (CNFs) as a single substance, even though experimental data have demonstrated significant differences in CNT toxicity. The unique combinations of physical and chemical characteristics cause variations in the observed dose-response-recovery that is substantial enough to consider them as a collection of different substances rather than a single substance. This paper presents an algorithm capable of grouping CNTs into toxicologically distinct clusters assisting in the identification of physicochemical differences between the clusters, and different proposed exposure limits for each of the CNT classes. Based on a dataset composed of peer-reviewed in vivo experimental studies in rodents, the CNT variants are divided into sub-groups based on their dose-response-recovery similarity and the Akaike Information Criterion (AIC) of the family of models. Results indicate the presence of 4 toxicologically unique CNT classes based on 5 toxicity endpoints selected. Certain physicochemical attributes vary significantly between clusters and are more likely to define the categories than others. The potency of the clusters is derived from their associated dose-response-recovery relationship parameters. The clusters with largest potency was found to be between 80 and 400 times more potent than the cluster with the lowest potency indicative of a large spread between the values across all the responses. The absence of key characterization data for some of the CNT variants analyzed in this study prevents the designation of physical characteristic based labels which could have assisted in identifying the key factors affecting the toxic potential of CNTs. The standardized potency of the 4 CNT clusters was compared, 2 clusters showed increased levels of immune response activity, 1 cluster only showed increased cell damage indicators and, 1 cluster displayed both elevated immune response and cell damage indicators based on the short-term endpoints measured in the test subjects.
All Science Journal Classification (ASJC) codes
- Computer Science Applications
- Health, Toxicology and Mutagenesis