toxbai

is a Data Source.

It is part of the NAMs collection.

ToxBAI is a desktop application for predicting the toxicity of chemicals with mechanistically explainable AI organized around the adverse outcome pathway (AOP) framework. From a chemical structure (CAS number or SMILES), it computes molecular descriptors and fingerprints and performs Active/Inactive binary classification across toxicity endpoints, reporting a confidence level, performance metrics (F1, recall, AUC), training dataset size, and an applicability-domain reliability flag for each prediction. It offers two families of models — apical-outcome (AO) prediction models trained on OECD Test Guideline data that map to regulatory endpoints, and mechanism prediction models trained on ToxCast high-throughput bioactivity data that correspond to molecular initiating events and key events; predicted MIEs and KEs are mapped to AOP networks in the OECD AOP-Wiki for mechanistic interpretation. ToxBAI is developed by the Environmental Systems Toxicology group at the University of Seoul and operated by ChemBAI, in collaboration with RTI International. It supports pre-screening of untested chemicals under the Next Generation Risk Assessment (NGRA) framework.

Product Summary

Products

From this Resource
ID Name URL Category Format Description
toxbai.app ToxBAI Prediction Application toxbai.com Graphical Interface http ToxBAI desktop application (Windows a...
toxbai.userguide ToxBAI User Guide TOXBAI_Userguide_beta_EN.pdf (11.6 MB) Documentation pdf ToxBAI user guide (beta, English), co...

Details

ToxBAI

Overview

ToxBAI is a desktop application for predicting chemical toxicity using mechanistically explainable AI built around the adverse outcome pathway (AOP) framework. Given a chemical structure (CAS number or SMILES), it computes molecular descriptors and fingerprints and classifies the chemical as Active or Inactive for a range of toxicity endpoints, reporting a confidence level, performance metrics (F1, recall, AUC), training set size, and an applicability-domain reliability flag for each prediction.

It provides two families of models:

  • AO prediction models, trained on OECD Test Guideline data, corresponding to regulatory apical endpoints.
  • Mechanism prediction models, trained on ToxCast high-throughput bioactivity data, corresponding to molecular initiating events (MIEs) and key events (KEs).

Predicted MIEs and KEs are mapped to AOP networks in the OECD AOP-Wiki, so a prediction can be interpreted at the biological-pathway level. Endpoint areas include endocrine disruption and developmental and reproductive toxicity (ED-DART), developmental neurotoxicity, inhalation toxicity, genotoxicity and cancer, and liver toxicity bioactivity.

ToxBAI is developed by the Environmental Systems Toxicology (EST) group at the University of Seoul and operated by ChemBAI, in collaboration with RTI International. It is designed to support pre-screening of untested chemicals under the Next Generation Risk Assessment (NGRA) framework.

Access

  • Download the Windows or macOS application from http://toxbai.com/. A Google sign-in and one-time acceptance of the terms are required.
  • The user guide is available as a PDF from the same site.

Version 1.0 provides the prediction toolbox; the AOP-based AI and NGRA 1.0 panels are shown as coming-soon placeholders. There is no public API or bulk data download.

Citation

A describing manuscript (Kim et al., 2026) is in preparation; no DOI is available yet.

Is this information incorrect or incomplete? Request an update.

Created: October 28, 2026 | Last modified: October 28, 2026