MutagenPred-GCNNs: A Graph Convolutional Neural Network-Based Classification Model for Mutagenicity Prediction with...

Area of Science:

• Computational chemistry
• Toxicology
• Drug discovery

Background:

• Identifying mutagenic compounds is crucial in early drug discovery.
• Predictive models linking compound structure to toxicological data are highly desirable.

Purpose of the Study:

• To develop a predictive model for compound mutagenicity using graph convolutional neural networks (GCNNs).
• To enable the identification of structural alerts and toxicophores within compounds.

Main Methods:

• Utilized an advanced graph convolutional neural network (GCNN) architecture.
• Extracted molecular representations to train predictive models.
• Validated model performance using fivefold cross-validation and external datasets.

Main Results:

• Achieved high performance metrics, including Area Under the Curve (AUC) up to 0.8782.
• Demonstrated accurate prediction of mutagenicity and identification of toxicophores like aromatic nitro and quinones.
• GCNNs effectively learned mutagenic features from molecular data.

Conclusions:

• Developed a highly predictive and interpretable mutagenicity classification model.
• The GCNN-based approach offers a powerful tool for assessing compound safety in drug discovery.
• Data-driven molecular representations learned by GCNNs enhance predictive toxicology.