Résumé:
Artificial intelligence is increasingly being utilized to all field and specially to enhance
occupational health within the pharmaceutical industry, where the rising potency of compounds
contributes to elevated exposure risks. Due to limited toxicity data for many emerging drug
candidates, traditional methods often fall short in accurately estimating safe handling
thresholds. In this study, an AI-driven approach has been developed to predict Occupational
Exposure Bands (OEBs) based on molecular structures. This method combines
cheminformatics descriptors and molecular fingerprints with deep learning techniques to
extract significant features, which are then classified using various machine learning
algorithms. The resulting models exhibit strong predictive performance, although challenges
remain in accurately identifying less-represented high-risk categories. Additionally, a practical
software tool was created to facilitate real-time OEB predictions and molecular visualization,
providing an accessible interface for researchers and safety professionals. Overall, this
approach offers an innovative solution for early hazard assessment and highlights the potential
of AI to improve workplace safety in pharmaceutical development.
