MIFNN: Molecular Information Feature Extraction and Fusion Deep Neural Network for Screening Potential Drugs

Area of Science:

• Computational chemistry
• Cheminformatics
• Artificial intelligence in drug discovery

Background:

• Molecular property prediction is crucial for efficient drug screening and reducing discovery costs.
• Deep learning approaches have shown promise in predicting molecular properties.
• Existing methods may lack comprehensive feature extraction and efficient information processing.

Purpose of the Study:

• To propose a novel deep learning model, the Molecular Information Fusion Neural Network (MIFNN), for improved molecular property prediction.
• To enhance the comprehensiveness of molecular feature extraction and information fusion.
• To boost classification accuracy and efficiency in drug discovery applications.

Main Methods:

• Utilized 1D-CNN for directed molecular information and 2D-CNN for Morgan fingerprints for comprehensive feature extraction.
• Fused 1D and 2D molecular features using directed message-passing to reduce information redundancy and enhance efficiency.
• Incorporated bidirectional long short-term memory and an attention module to refine molecular features and improve classification.
• Optimized a support vector machine using particle swarm optimization.

Main Results:

• The MIFNN model was evaluated on eight public datasets, demonstrating robust performance.
• Ablation experiments confirmed the effectiveness of individual modules within the MIFNN architecture.
• The model achieved a maximum improvement of 14% on the ToxCast dataset compared to baseline models.
• MIFNN exhibited stable performance across most datasets, outperforming previous models.

Conclusions:

• The proposed MIFNN model offers superior performance for molecular property prediction compared to existing methods.
• The fusion of multi-dimensional molecular features and advanced deep learning modules significantly enhances predictive accuracy.
• MIFNN represents a promising advancement in computational drug discovery, facilitating more efficient screening and cost reduction.