Computational approaches have recently gained a lot of traction in the field of read-across. Previously we developed our generalized read-across (GenRA) approach, which utilizes in vitro bioactivity data in conjunction with chemical descriptor information to derive local validity domains to facilitate read-across prediction. Predictions are made using the similarity weighted average of nearest neighbors. GenRA has been used to predict up to 574 different apical effects as a binary call from repeat-dose toxicity studies available in ToxRefDB. Here we evaluate the application of GenRA for quantitative predictions, specifically using a large dataset of rat oral acute LD50 toxicity data. These data were constructed by mining and merging a number of publicly available resources and comprise 21,210 unique LD50 values for 15,698 discrete chemicals. We used 9293 chemicals for which structures were available in the US EPA CompTox Chemicals Dashboard to make GenRA predictions of LD50 values. GenRA predictions used Morgan chemical fingerprints with a minimum Jaccard similarity threshold of 0.5 and a maximum of 10 nearest neighbors over the entire dataset. We evaluated confidence in GenRA predictions using four-fold cross-validation testing producing r<sup>2</sup> values 0.56±0.3. These findings show the utility of GenRA for making quantitative predictions of toxicity.

Case study in quantitative GenRA predictions using acute oral toxicity

2019 Spring National Meeting

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.