Published September 25, 2018

As part of its ongoing computational toxicology research, EPA is developing faster and improved approaches to evaluate chemicals for potential health effects.  One commonly applied approach is known as chemical read-across. Read-across uses information about how a chemical with known data behaves to make a prediction about the behavior of another chemical that is “similar” but does not have as much data. Current read-across, while cost-effective, relies on a subjective assessment, which leads to varying predictions and justifications depending on who undertakes and evaluates the assessment.

To reduce uncertainties and develop a more objective approach, EPA researchers have developed an automated read-across tool called Generalized Read-Across (GenRA), and added it to the newest version of the EPA Computational Toxicology Dashboard. The goal of GenRA is to encode as many expert considerations used within current read-across approaches as possible and combine these with data-driven approaches to transition read-across towards a more systematic and data-based method of making predictions.

EPA chemist Dr. Grace Patlewicz says it was this uncertainty that motivated the development of GenRA. "You don't actually know if you've been successful at using read-across to help predict chemical toxicity because it's a judgement call based on one person versus the next. That subjectivity is something we were trying to move away from." Patlewicz says.

Since toxicologists and risk assessors are already familiar with read-across, EPA researchers saw value in creating a tool that that was aligned with the current read-across workflow but which addressed uncertainty using data analysis methods in what they call a “harmonized-hybrid workflow.”

In its current form, GenRA lets users find analogues, or chemicals that are similar to their target chemical, based on chemical structural similarity. The user can then select which analogues they want to carry forward into the GenRA prediction by exploring the consistency and concordance of the underlying experimental data for those analogues. Next, the tool predicts toxicity effects of specific repeated dose studies. Then, a plot with these outcomes is generated based on a similarity-weighted activity of the analogue chemicals the user selected. Finally, the user is presented with a data matrix view showing whether a chemical is predicted to be toxic (yes or no) for a chosen set of toxicity endpoints, with a quantitative measure of uncertainty.

The team is also comparing chemicals based on other similarity contexts, such as physicochemical characteristics or metabolic similarity, as well as extending the approach to make quantitative predictions of toxicity.

Patlewicz thinks incorporating other contexts and similarity measures will refine GenRA to make better toxicity predictions, fulfilling the goal of creating a read-across method capable of assessing thousands of chemicals that currently lack toxicity data.

“That's the direction that we're going in,” Patlewicz says. “Recognizing where we are and trying to move towards something a little bit more objective, showing how aspects of the current read-across workflow could be refined.”

Learn more at: https://comptox.epa.gov