AMPred-CNN: Ames mutagenicity prediction model based on convolutional neural networks.

Area of Science:

• Computational toxicology
• Cheminformatics
• Machine learning for drug discovery

Background:

• Mutagenicity assessment is crucial for evaluating chemical and pharmaceutical safety.
• Computational models offer efficient and cost-effective toxicity predictions.
• Convolutional Neural Networks (CNNs) excel at feature extraction from image-like data.

Purpose of the Study:

• To introduce AMPred-CNN, a novel Ames mutagenicity prediction model utilizing CNNs.
• To leverage molecular structures as images for enhanced feature extraction.
• To evaluate AMPred-CNN's performance against traditional and advanced models.

Main Methods:

• Development of AMPred-CNN using Convolutional Neural Networks (CNNs).
• Utilizing molecular structures represented as images for input.
• Model training and evaluation on the Hansen et al. benchmark mutagenicity dataset.
• Comparative analysis with traditional machine learning (ML) models and recent deep learning (DL) models.

Main Results:

• AMPred-CNN demonstrated superior performance compared to traditional ML models across various metrics (accuracy, AUC, F1 score, MCC, sensitivity, specificity).
• Benchmarking against seven recent ML and DL models showed consistently superior results for AMPred-CNN.
• Achieved an impressive Area Under the Curve (AUC) of 0.954.

Conclusions:

• CNNs are highly effective for advancing mutagenicity prediction accuracy.
• AMPred-CNN represents a significant improvement in computational toxicology.
• The model shows promise for broader applications in drug development and safety evaluation.