Artificial Intelligence in Toxicology: New Breakthroughs in Assessing Compound Toxicity

Strategic Insights -

Researchers from MicroDiscovery GmbH, in collaboration with several international colleagues, have made significant progress in utilizing artificial intelligence (AI) to identify and assess the toxicity of compounds. By leveraging three distinct methods for automatic information extraction from unstructured text – text mining, word embeddings, and large language models – the team has developed a comprehensive approach to calculate a hepatotoxicity score for over 50,000 compounds.

This breakthrough has the potential to revolutionize the field of toxicology, enabling scientists to rapidly identify potential hepatoxicants and streamline the process of synthesizing vast amounts of published information. The researchers' findings indicate that combining these methods further improves performance, yielding an Area Under the Curve (AUC) of 0.87 in Drug-Induced Liver Injury (DILI) validation.

Key Takeaways:

  • The research team employed three distinct methods for automatic information extraction from unstructured text: text mining, word embeddings, and large language models.
  • Text mining achieved an Area Under the Curve (AUC) of 0.8 in DILI validation, while large language models performed better with an AUC of 0.85.
  • Combining these methods yielded an AUC of 0.87 in DILI validation, indicating improved performance.
  • The study assessed the hepatotoxicity of over 50,000 compounds using a comprehensive approach.
  • The researchers developed a method to automatically identify potential hepatoxicants from over 50,000 compounds using the wealth of scientific publications and knowledge.
  • The study utilized a use case on Drug-Induced Liver Injury (DILI) to evaluate the performance of the different methods.
  • The research team included Chris Bauer, Long Tran Duc Dang, Twan van den Beucken, Johannes Schuchhardt, and Ralf Herwig as authors.
  • The study's findings have significant potential to revolutionize the field of toxicology.

Statistics:

  • The research team assessed the hepatotoxicity of over 50,000 compounds.
  • The Area Under the Curve (AUC) for text mining in DILI validation was 0.8.
  • The AUC for large language models in DILI validation was 0.85.
  • The AUC for the combined method in DILI validation was 0.87.
  • The study utilized over 50,000 compounds in its analysis.
  • The research team developed a method to automate the identification of potential hepatoxicants from over 50,000 compounds.

Sources:

  • Systematic analysis of hepatotoxicity: combining literature mining and AI language models, Frontiers in Artificial Intelligence, 2025, 8. (https://doi-org.sdpl.idm.oclc.org/10.3389/frai.2025.1561292)
  • NewsRx. Research from MicroDiscovery GmbH Provides New Data on Artificial Intelligence (Systematic analysis of hepatotoxicity: combining literature mining and AI language models). Gastroenterology Week. August 4, 2025; p 465.