Text

theses

Chemical toxicity assessment is important to public health since numerous chemicals are being used daily and the chemical exposed to human beings may cause potential toxic effects. Traditional methods for toxicity test of chemicals, such as standard rodent models, are expensive and time consuming. Along with the vibrant and rapid progress of chemical synthesis and biological screening technologies (e.g. high-throughput screening), immense in vitro toxicity data are generated daily and most of these data are available to the public through various data sharing project. The enormous toxicity data possess the intrinsic “five Vs” characteristics of big data (i.e. volume, velocity, variety, veracity and value), and moved traditional toxicology into a “big data” era. However, the relevance between these fast accumulating in vitro toxicity data with the immediate human toxicity effect is obscure. Computational modeling, originally as an alternative method to animal models, showed promising ability to bridge the public toxicity big data to potential chemical toxicity effects in human beings. Thus, it is necessary to develop novel computational models to answer the challenges brought by big data. In this dissertation, new computational models and associated modeling approaches were described for toxicity assessments of chemicals using public big data. First, a method for the identification of uncertainty in the training data used for quantitative structure−activity relationship (QSAR) modeling was developed, which addressed potential issues relevant to veracity of toxicity data. Second, a hybrid read-across method was developed, which focused on handling the data obtained from various resources (i.e. the variety of toxicity big data). A hybrid read-across study, which were based on the combination of chemical descriptors and biological data, showed better predictivity than traditional read-across results that were based on chemical similarity. Last, novel mechanism- driven read-across approach was developed specifically for chemical hepatotoxicity evaluations. A virtual adverse outcome pathway (vAOP) modeling tool were developed and validated using a large hepatotoxicity database. This read-across study showed promising applicability to the prediction of new compounds for their hepatotoxicity and answered the current five Vs challenges of toxicity big data.

ETD_10950

Ph.D.

Includes bibliographical references

doi:10.7282/t3-9vkr-a015

ETD doctoral

The author owns the copyright to this work.