Building Quantitative Prediction Models for Tissue Residue of Two Explosives Compounds in Earthworms from Microarray Gene Expression Data

Abstract

Soil contamination near munitions plants and testing grounds is a serious environmental concern that can result in the formation of tissue chemical residue in exposed animals. Quantitative prediction of tissue residue still represents a challenging task despite long-term interest and pursuit, as tissue residue formation is the result of many dynamic processes including uptake, transformation, and assimilation. The availability of high-dimensional microarray gene expression data presents a new opportunity for computational predictive modeling of tissue residue from changes in expression profile. Here we analyzed a 240-sample data set with measurements of transcriptomic-wide gene expression and tissue residue of two chemicals, 2,4,6-trinitrotoluene (TNT) and 1,3,5-trinitro-1,3,5-triazacyclohexane (RDX), in the earthworm Eisenia fetida. We applied two different computational approaches, LASSO (Least Absolute Shrinkage and Selection Operator) and RF (Random Forest), to identify predictor genes and built predictive models. Each approach was tested alone and in combination with a prior variable selection procedure that involved the Wilcoxon rank-sum test and HOPACH (Hierarchical Ordered Partitioning And Collapsing Hybrid). Model evaluation results suggest that LASSO was the best performer of minimum complexity on the TNT data set, whereas the combined Wilcoxon-HOPACH-RF approach achieved the highest prediction accuracy on the RDX data set. Our models separately identified two small sets of ca. 30 predictor genes for RDX and TNT. We have demonstrated that both LASSO and RF are powerful tools for quantitative prediction of tissue residue. They also leave more unknown than explained, however, allowing room for improvement with other computational methods and extension to mixture contamination scenarios.

A Meta‐Analysis of Carbon Nanotube Pulmonary Toxicity Studies—How Physical Dimensions and Impurities Affect the Toxicity of Carbon Nanotubes

“…RT and RF models have been employed in generating quantitative predictive models for biological activity of different chemical compounds including toxicity and bioaccumulation, applications similar to this study of physical and chemical characteristics associated with CNT toxicity. Investigators have also used RF models to identify important epidemiological and phenotypic differences in salmonella strains .…”

Wildlife Toxicity Assessment for 1,3,5-Trinitrohexahydro-1,3,5-Triazine (RDX)

“…A range of bioinformatic approaches have been used to investigate the impact of RDX toxicity in a range of species including rats [55], quail [31], fish [56], and earthworms [57][58][59][60][61] with a view to elucidating mechanisms of toxicity and developing predictive models for enhanced monitoring. While these exploratory studies reveal a consistent range of responses to RDX across species, the range of differentially expressed genes or enriched pathways still represents a somewhat complex descriptive toxicology, with the results highly dependent on the choice of statistical methods.…”

Binding of RDX to Cell Wall Components ofPinus sylvestrisandPicea glaucaand Three-Year Mineralisation Study of Tissue-Associated RDX Residues

“…Standard procedures, described elsewhere (Mezzari et al 2004;Gong et al 2012), were applied for the determination of 14 C. Briefly, 14 C-containing solid organic material was combusted oxidatively in a biological oxidizer (OX 500, Zinsser Analytik, Frankfurt/M., Germany). The resulting 14 CO 2 was trapped in an Oxysolve C-400 scintillation cocktail.…”