0
Research Papers

Nanoparticle Characteristic Interaction Effects on Pulmonary Toxicity: A Random Forest Modeling Framework to Compare Risks of Nanomaterial Variants

[+] Author and Article Information
Jeremy M. Gernand

Department of Energy and Mineral Engineering,
The Pennsylvania State University,
110 Hosler Bldg., University Park, PA 16802
e-mail: jmgernand@psu.edu

Elizabeth A. Casman

Engineering and Public Policy Department,
Carnegie Mellon University,
5000 Forbes Ave., Pittsburgh, PA 15213
e-mail: casman@andrew.cmu.edu

1Corresponding author.

Manuscript received February 18, 2015; final manuscript received July 29, 2015; published online January 4, 2016. Assoc. Editor: Chimba Mkandawire.

ASME J. Risk Uncertainty Part B 2(2), 021002 (Jan 04, 2016) (13 pages) Paper No: RISK-15-1022; doi: 10.1115/1.4031216 History: Received February 18, 2015; Accepted July 31, 2015

Due to their unique physicochemical properties, nanomaterials have the potential to interact with living organisms in novel ways. Nanomaterial variants are too numerous to be screened for toxicity individually by traditional animal testing. Existing data on the toxicity of inhaled nanomaterials in animal models are sparse in comparison to the number of potential factors that may affect toxicity. This paper presents meta-analysis-based risk models developed with the machine-learning technique, random forests (RFs), to determine the relative contribution of different physical and chemical attributes on observed toxicity. The findings from this analysis indicate that carbon nanotube (CNT) impurities explain at most 30% of the variance in pulmonary toxicity as measured by polymorphonuclear neutrophils (PMNs) count. Titanium dioxide nanoparticle size and aggregation affected the observed toxic response by less than 10%. Differences in observed effects for a group of metal oxide nanoparticles associated with differences in Gibbs free energy on lactate dehydrogenase (LDH) concentrations amount to only 4% to the total variance.

Copyright © 2016 by ASME
Your Session has timed out. Please sign back in to continue.

References

Figures

Grahic Jump Location
Fig. 1

Example of a single-branch RT with two leaf nodes displaying the predicted mean output (Y) and standard deviation at each node along with the sample size (N)

Grahic Jump Location
Fig. 2

RF model error as a function of trees included in model for the prediction of BAL LDH following pulmonary exposure to titanium dioxide nanoparticles

Grahic Jump Location
Fig. 3

(a) Change in neutrophil count in BAL fluid following pulmonary exposure to CNTs, (b) change in LDH in BAL fluid following pulmonary exposure to CNTs, (c) change in LDH in BAL fluid following exposure to titanium dioxide nanoparticles, and (d) change in total protein in BAL fluid following exposure to titanium dioxide nanoparticles

Grahic Jump Location
Fig. 4

Effects of pulmonary exposure to CNTs at three dose levels, and all values of nanotube length and diameter: minimum dose is 2  μg/kg; median dose is 3250  μg/kg; maximum dose is 6500  μg/kg. (a) Change in LDH in BAL fluid following exposure. (b) Change in neutrophils count in BAL fluid following exposure. Values other than dose, length, and diameter, such as recovery period, and % cobalt impurity are held constant at their median reported values. These results suggest that larger diameter CNTs (multiwalled CNTs) produce a significantly increased immune response (PMN counts), but only a mildly increased LDH concentration.

Grahic Jump Location
Fig. 5

Effects of pulmonary exposure to titanium dioxide nanoparticles based on changes in dose, aggregate diameter (MMAD), and purity. (a) Changes in LDH in BAL fluid and (b) changes in total protein concentration in BAL fluid. Other variables in the model are held constant at their median values. The minimum dose is 35  μg/kg. The median dose is 1.8×106  μg/kg. The maximum dose is 3.5×106  μg/kg. These results indicate that increasing purity is associated with a mildly decreasing LDH concentration, but has little impact on total protein concentration. The size of particle aggregates appears to have negligible effect for either measure.

Grahic Jump Location
Fig. 6

Effects of pulmonary exposure to metal oxide nanoparticles including titanium dioxide, zinc oxide, magnesium oxide, and silicon dioxide, based on changes in (a) aggregation (MMAD) and the Gibbs free energy, and primary particle size and specific surface area (b). The minimum dose is 300  μg/kg, the median dose is 8000  μg/kg, and the maximum dose is 16,000  μg/kg. These plots indicate that changes in total dose by mass affect the observed toxicity to a much greater degree than any effects from size or chemical factors.

Grahic Jump Location
Fig. 7

Changes in RF-model-predicted BAL neutrophils count following exposure to CNTs as a function of changes in total dose and the dose of cobalt, a common toxic impurity (up to 0.53% by weight of total CNTs). This suggests that Co and total CNTs both independently contribute to higher neutrophils count.

Grahic Jump Location
Fig. 8

Changes in RF-model-predicted BAL LDH following exposure to CNTs as a function of changes in total dose and the dose of cobalt, a common toxic impurity (up to 0.53% by weight of total CNTs). This suggests that total dose is much more important than Co content for increasing LDH concentration.

Grahic Jump Location
Fig. 9

Changes in RF-model-predicted BAL neutrophils count following exposure to CNTs as a function of changes in total dose and aggregation (MMAD). This suggests that aggregation only has a small effect on neutrophils count as compared to total dose, and also that low to moderate doses are relatively similar in response.

Grahic Jump Location
Fig. 10

Changes in RF-model-predicted BAL LDH following exposure to CNTs as a function of changes in total dose and aggregation (MMAD). This suggests both that total dose is a more important predictor of LDH than aggregation, but also that less aggregation can increase LDH as well.

Grahic Jump Location
Fig. 11

Changes in RF-model-predicted BAL total protein following exposure to titanium dioxide nanoparticles as a function of changes in total dose and aggregation (MMAD). This suggests that low aggregation levels are more toxic than higher ones, and that recovery time is not an important factor, but the scale of these differences is small overall.

Grahic Jump Location
Fig. 12

Changes in RF-model-predicted BAL LDH following exposure to titanium dioxide nanoparticles as a function of changes in total dose and aggregation (MMAD). This suggests that in terms of predicting LDH response, neither recovery time nor aggregation is consistently detrimental or beneficial, so more data would be required.

Grahic Jump Location
Fig. 13

Changes in RF-model-predicted BAL LDH following exposure to metal oxide nanoparticles (TiO2, MgO, ZnO, SiO2) as a function of changes in aggregation (MMAD), and purity. Based on the scale, there is little difference in toxicity across this range of variables indicating that these factors play little role in toxicity.

Grahic Jump Location
Fig. 14

Changes in RF-model-predicted BAL LDH following exposure to metal oxide nanoparticles (TiO2, MgO, ZnO, SiO2) as a function of changes in aggregation (MMAD), and Gibbs Free Energy, a descriptor of the chemical energy available in the metal oxide compound. This suggests that aggregation is much less important than differences in chemical makeup (other results, see Fig. 5, indicate that Gibbs Free Energy is much less important than total dose).

Grahic Jump Location
Fig. 15

Changes in RF-model-predicted BAL LDH following exposure to metal oxide nanoparticles (TiO2, MgO, ZnO, SiO2) as a function of total dose and recovery period. This indicates that total dose dominates the change in LDH due to longer recovery periods.

Grahic Jump Location
Fig. 16

Changes in RF-model-predicted BAL LDH following exposure to titanium dioxide nanoparticles as a function of total dose and average particle size. This indicates that very small TiO2 nanoparticles are more toxic than those in most of the possible range of sizes.

Grahic Jump Location
Fig. 17

RF model variable importance as measured by variance reduction attributable to each variable for the prediction of BAL neutrophils count following exposure to CNTs

Grahic Jump Location
Fig. 18

RF model variable importance as measured by variance reduction attributable to each variable for the prediction of BAL LDH count following exposure to CNTs

Grahic Jump Location
Fig. 19

RF model variable importance as measured by variance reduction attributable to each variable for the prediction of BAL LDH count following exposure to titanium dioxide nanoparticles

Grahic Jump Location
Fig. 20

RF model variable importance as measured by variance reduction attributable to each variable for the prediction of BAL total protein count following exposure to titanium dioxide nanoparticles

Grahic Jump Location
Fig. 21

RF model variable importance as measured by variance reduction attributable to each variable for the prediction of BAL LDH count following exposure to metal oxide nanoparticles include titanium dioxide, magnesium oxide, and silicon dioxide

Grahic Jump Location
Fig. 22

RF model error as a function of trees included in model for prediction of BAL neutrophils following pulmonary exposure to CNTs

Grahic Jump Location
Fig. 23

RF model error as a function of trees included in model for prediction of BAL LDH following pulmonary exposure to CNTs

Grahic Jump Location
Fig. 24

RF model error as a function of trees included in model for prediction of BAL total protein following pulmonary exposure to titanium dioxide nanoparticles

Grahic Jump Location
Fig. 25

RF model error as a function of trees included in model for prediction of BAL LDH following pulmonary exposure to metal oxide nanoparticles including titanium dioxide, magnesium oxide, silicon dioxide, and zinc oxide

Tables

Errata

Discussions

Some tools below are only available to our subscribers or users with an online account.

Related Content

Customize your page view by dragging and repositioning the boxes below.

Articles from Part A: Civil Engineering
Related eBook Content
Topic Collections

Sorry! You do not have access to this content. For assistance or to subscribe, please contact us:

  • TELEPHONE: 1-800-843-2763 (Toll-free in the USA)
  • EMAIL: asmedigitalcollection@asme.org
Sign In