Deep Learning Model for Identifying Critical Structural Motifs in Potential Endocrine Disruptors

Area of Science:

• Computational chemistry and toxicology
• Artificial intelligence in drug discovery and safety assessment

Background:

• Endocrine-disrupting chemicals (EDCs) pose significant environmental and health risks.
• Identifying structural features responsible for endocrine disruption is crucial for chemical safety.
• Current methods for assessing endocrine disruption potential can be time-consuming and resource-intensive.

Purpose of the Study:

• To develop a deep neural network-based toolkit for identifying structural motifs contributing to endocrine disruption.
• To enable rapid, virtual assessment of a synthetic chemical's endocrine disruption potential.
• To pinpoint specific structural alerts and their chemical environments responsible for endocrine activity.

Main Methods:

• Development of a multilabel/multioutput classification model combining Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) architectures.
• Utilization of an active learning framework integrating diverse data sources for model training.
• Application of Class Activation Maps (CAMs) to visualize and interpret model predictions, identifying key structural features.

Main Results:

• The developed toolkit accurately assesses endocrine disruption potential using simplified molecular input line entry system (SMILES) representations.
• The model successfully identifies critical structural alerts and their associated chemical environments.
• Class activation maps provide interpretable insights into the molecular features driving endocrine disruption predictions.

Conclusions:

• The deep neural network toolkit offers an efficient and accurate method for predicting endocrine disruption potential.
• This approach aids in identifying hazardous structural motifs, facilitating the design of safer chemicals.
• The toolkit advances computational toxicology by integrating AI with interpretable methods for chemical safety assessment.