Adapting computer science technologies to bridge critical gaps in proposed Tox21 risk assessment programs

Graduation date: 2015In 2004 the National Toxicology Program published an article describing their intent to pursue toxicology testing and risk assessment practices solely reliant on high throughput in vitro datasets and in silico modeling for dose-response assessments. Support and contributions form regulatory agencies and a significant portion of the life sciences community soon led to widespread commitment of pursuing a "21st century" or a "Tox21" risk assessment program. As of 2014, the envisioned program is still far from realized, as the difficulty in developing such a monumental change in risk assessment practice that can withstand scrutiny under the precautionary principle has proved far more difficult than originally anticipated. We developed several novel adaptations of existing computer science technologies for the purposes of establishing necessary components of a Tox21 program and identifying current critical research gaps in proposed Tox21 risk assessment programs.\ud A fuzzy neural network model was created to quantitatively predict changes in Cyp1b1 gene expression following complex PAH mixture-mediated perturbations of the Cyp1b1 gene regulatory network. Predictions were within one log₂ fold change unit for four out five treatment groups, demonstrating the feasibility of predicting responses following complex co-exposures in a biological pathway of interest, similar to a mode of action dose-response paradigm. Predictions were dependent on the feedback component of the Cyp1b1 gene regulatory network (the aryl hrydocarbon receptor repressor) for model predictions, demonstrating the need to include inhibitor and activator components of mode of action pathways in datasets used for Tox21 risk assessments. Model cross validation studies and summary statistics, including model predictions, were automated, demonstrating the feasibility of an automated high throughput approach to dose-response assessment.\ud Automated processes were also developed for atmospheric models of hourly fine particulate matter, coarse particulate matter, and ozone as new meteorological and pollutant sampling measurements were posted online by the Oregon Department of Environmental Quality. Air quality predictions were combined with smartphone location measurements to predict air quality conditions at smartphone locations and communicate predictions with smartphone users in real time, demonstrating the feasibility of incorporating individual, personalized measurements in an automated exposure assessment program as new individual and regional measurements become available. The software interface between smartphones and atmospheric models was developed with a high degree of modularity, allowing for communications between multiple smartphone platforms, and multiple atmospheric modeling methodologies with varying degrees of temporal and spatial resolutions, demonstrating the feasibility of developing a Tox21 risk assessment structure that can update dynamically as technologies are improved and new technologies become available.\ud Air quality communications were enhanced through interactive components of electronic media, specifically through allowing users to set customized warning levels and select between spatial, temporal, or summary statistics displays of predicted air quality conditions. \ud Affect heuristic and economic behavior theory analysis of webmap, smartphone, and augmented reality displays developed for air quality risk communications identified several potential improvements to enhance risk perception/ risk communication efforts. Analysis supported the utility of interactive electronic media in customizing risk communications for enhanced risk perceptions, stakeholder input, and transparency in a future Tox21 program.\ud Collectively these studies demonstrate that the realization of a Tox21 program is limited more by our current lack of knowledge and ability to design, utilize, and evaluate modern technology applications and methodologies than by the underlying technologies' current state