Integration of Biodescriptors and Chemodescriptors for Predictive Toxicology: A Mathematical / Computational Approach

Abstract

i) Attempts have been made to develop quantitative biodescriptors to characterize the effects of chemicals on the proteomics patterns of cells. Applications of biodescriptors to proteomics patterns derived from normal liver cells and cells exposed to peroxisome proliferators showed that quantitative descriptors defined for D/D matrices are capable of rationalizing effects of different types of peroxisome proliferators on the cellular proteome. Such descriptors may find application in predicting the biochemical effects of toxicants from their proteomics patterns. ii) A novel approach to biodescriptor formulation has been taken, using the concept of partial ordering. The embedded graphs derived thereby have been used to develop new types of invariants for the quantification of proteomics maps. Biodescriptors developed and reported in the above two manuscripts show reasonable power of discriminating the proteomics patterns resulting from different toxicants. Biodescriptors already developed in this project and those currently under development will provide a battery of such parameters for predictive toxicology. When a certain number of (say n) of such descriptors are calculated for a proteomics map, the map is characterized by a vector consisting of n entries. Elements of such vectors may be used in: a) predicting toxicity using statistical methods such as PCR, PLS, RR or non-linear methods such as neural networks; b) classifying chemicals into various subsets; or c) quantifying the similarity / dissimilarity of toxicants. In light of our earlier study on the prediction of the mode of action (MOAs) of toxicants from chemodescriptors, it is realistic to speculate that the battery of biodescriptors will be useful in predicting the MOAs of toxicants. Chemodescriptors: i) Chemodescriptors calculated by POLLY and Molconn-Z have been used to predict the cell level toxicity data of halocarbons determined at the Wright Patterson AFB laboratory by Kevin Geiss (unpublished results). The high quality of these models indicate that similar models using calculated chemodescriptors may find application in the prediction of cellular toxicity arising from exposure to pollutants and xenobiotics. ii) In collaboration with Dr. Hawkins, PCR, RR and PLS analyses have been applied-to the prediction of the property / activity / toxicity of various groups of chemicals from their structural descriptors, vis., topostructural indices, topochemical indices, 3-D or shape parameters, and semi-empirical quantum chemical descriptors. This research has resulted in the formulation of robust and useful QSTR methods. iii) Successful quantitative structure-toxicity relationship (QSTR) models have been developed for the exposure assessment of volatile organic chemicals (VOCs), such as halocarbons, using chemodescriptors with the ridge regression statistical technique. These models will be useful in the physiologically-based pharmacokinetic modeling of VOCs. iv) Novel molecular shape descriptors have been developed and applied to predicting properties dependant on molecular shape. These new molecular shape parameters will be useful in QSAR/QSTR studies where molecular shape is a critical determinant of ligand-biotarget interaction in the cellular milieu. v) New methods have been developed to characterize DNA sequences and their modifications as a result of exposure to toxicants using matrix invariants. Such invariants were defined from newly formulated matrices used to represent macromolecular sequences in a condensed manner. These invariants will be useful tools for research in genomics and bioinformatics.