High Throughput Chemical Toxicity Screening As a Bayesian Signaling Game

Lyle D. Burgoon, Ph.D.
US Army Engineer Research and Development Center

Uterotrophic Assay

• too expensive
• take too long
• they are less relevant to humans or other species than desired
• there are ethical constraints

Over past decade we've seen recent advances in alternative to animal assay technologies and a push by national and international regulatory bodies to adopt these alternative approaches

Today we are seeing interest by US and other governments to make decisions using these in vitro high throughput screening tests and computational models

Molecular tests, quantitative structure activity relationship models, computational molecular binding assays

Decision Model:

• Go: No further testing -- take to production
• No-Go: Kill or further testing is required before production

Datasets Used:

• US EPA ToxCast
• In vitro high throughput screening
• Used the data provider's "hit" calls (active in the assay)
• US NICEATM Uterotrophic Database
• Highly curated

11 estrogen receptor assays that need to be integrated somehow

Summary/Conclusions:

• The Bayesian Signaling game is a flexible approach for using HTS data to make Go/No-Go decisions
• Data integration using the Bayesian Network approach identified potential problematic assays
• Combining the Bayesian Signaling game with the Bayesian Network helped identify that we needed to improve our data integration strategy

Next Steps:

• Generate own hit calls using comparison to estrogen
• Package this approach up as much as possible into an R package