There is a strong need to develop New Alternative Methods (NAMs) to reduce animal testing of chemical, cosmetic, and pharmaceutical products to evaluate chemical toxicity. “6-pack” battery of regulatory assays (acute oral toxicity, acute dermal toxicity, acute inhalation toxicity, skin irritation and corrosion, eye irritation and corrosion, and skin sensitization) is a collection of tests that chemical products must go through to achieve regulatory approval. Computational approaches that can accurately estimate the results of the experimental testing can provide a powerful alternative to in vivo investigations. Previously, both our group and several other groups have developed models for some of these endpoints but using limited data or, in some cases, lacking rigor in both curation of the reported data and model validation strategies. This project addresses these deficiencies. We recently formed Predictive, LLC, to enable the development and distribution of commercial and regulatory strength models to predict important toxicity endpoints. In this Phase I STTR application, we intend to produce rigorously validated models of all “6-pack” assays, transfer these models to Predictive, LLC, and integrate these models into a software product termed STopTox (Systemic and Topical Toxicity) Predictor. We will achieve this objective by focusing on the following Specific Aims. Specific Aim 1. Develop advanced models for the “6-pack” battery of tests. We will ingest new data and develop new consensus models using multiple types of descriptors and advanced modeling techniques, including deep learning methods. We will also generate a Bayesian model applying individual predictions of each unique model as descriptors, which could assess if a compound would be active in any of the 6-pack tests. Specific Aim 2: Model interpretation and elucidation of adverse outcomes pathways (AOPs.) We will enable protocols and tools for model interpretation, which is an important part of regulatory decision support, both in terms of pf chemical features responsible for toxicity, and respective AOPs. Predictive probability maps will be implemented as a graphical visualization of the predicted fragment contribution, allowing the user to interpret the prediction and design safer compounds. In a parallel effort, we will work on the issue of AOPs, which is very important for a mechanistic understanding of toxicity mechanisms and regulatory acceptance of new chemicals. Specific Aim 3: STopTox platform development. Predictive, LLC, will implement all models in a software that will run both locally standalone and on a secure web portal. Testing will be done both internally and by external users. Predictions for individual models, the smart-consensus Bayesian models, as well as predicted fragment contributions, will be displayed on the screen and the user will be able to download a report with the results and a summary of characteristics of the models and instructions to help interpret the results. The ultimate objective of this proposal is to leverage public data knowledge on compounds tested in “6-pack” regulatory assays by creating a software platform (STopTox) to be commercialized as a service or licensed to commercial users.

Computational models provide powerful alternative to experimental toxicity studies that rely on the use of animals. Today, large collections of experimental data on chemical toxicity are available, which enables the development of computational toxicity models. Using highly curated data, this project will develop new computational models to predict systemic and topical toxicity of chemicals integrated into user-friendly STopTox package.