Toxicology-Prediction

A model that uses the Tox21 dataset [1] to attempt to find a relationship between molecular structure and relative toxicity, along with a Streamlit web app for interaction.

Toxicology Project Secondary Image

View GitHub for code


💡 Inspiration

A study from the American Society for Biochemistry and Molecular Biology (ASBMB) reports that:

  • 90% of drugs fail clinical testing
  • 30% of failures are due to molecular toxicity [2]

These failures result in billions of dollars in losses for pharmaceutical companies.

Our project addresses this challenge by:

  • Developing a machine learning model that predicts toxicity directly from molecular structure
  • Using Structure-Activity Relationship (SAR) principles to justify predictions based on similarity to known toxic compounds
  • Building an interactive Streamlit app to make the model accessible for exploration and teaching

⚛️ By combining machine learning with chemistry fundamentals, this project explores how AI can help reduce drug failure rates, save costs, and accelerate pharmaceutical research.


📚 References

[1] National Center for Advancing Translational Sciences, “Toxicology in the 21st Century (Tox21)”, [Online]. Available: https://tox21.gov. Accessed: Feb. 20, 2025.

[2] D. Sun, “90% of drugs fail clinical trials”, ASBMB Today, Mar. 12, 2022. [Online]. Available: https://www.asbmb.org/asbmb-today/opinions/031222/90-of-drugs-fail-clinical-trials. Accessed: Feb. 20, 2025.