Supervised Machine Learning Approaches Predict and Characterize Nanomaterial Exposures: MWCNT Markers in Lung Lavage Fluid.

Globally, carbon nanotubes (CNT) make up 30% of the total engineered nanomaterial market. Within that 30%, multi-walled carbon nanotubes (MWCNT) make up 94% of the total. Recent experimental evidence points towards significant pulmonary toxicity of MWCNTs such as inflammation, sub-pleural fibrosis and granuloma formation, associated with CNTs. Although numerous studies explore the adverse potential of various CNTs, their comparability is often limited. This is due to differences in administered dose, physico-chemical characteristics (e.g. agglomeration/aggregation state, metal impurities, stiffness, length) of the CNTs studied, exposure methods employed, as well as the differences in the end points monitored. In this study, we attempted to address the problem of identifying protein markers consistent across different MWCNT studies through the application of a sparse supervised classification methods. A panel of proteins measured in bronchoalveolar lavage collected from mice at various post-exposure time points and concentrations exposed to two different pristine or as-produced MWCNT, their polymer coated counterparts, or a well-studied reference material, MWCNT-7, were analyzed. The main objective was to take advantage of the power of sparse classification methods in identifying a small number of highly predictive and correlated markers (4 to 7, out of a panel of 52 proteins) that can distinguish exposure to MWCNT and/or be attributable to MWCNT toxicity in mice. Using this approach, we identified a small subset of proteins clearly distinguishing each exposure. MDC/CCL22, in particular, was associated with various MWCNT exposures and was independent of exposure route tested i.e., oropharyngeal aspiration versus inhalation exposure. The approaches presented in this study could enable comparison not only within a class of engineered nanomaterials but between various classes of nanomaterials. This study thus serves as a "proof of concept" that can be expanded to future nanomaterial risk profiling studies by informing decisions related to dose- and time-response relationships and to generate relevant experimental conditions.