PSO-FeatureFusion: a general framework for fusing heterogeneous features via particle swarm optimization.

Area of Science:

• Bioinformatics
• Computational Biology
• Data Integration

Background:

• Integrating heterogeneous biological data is crucial for understanding complex relationships between drugs, diseases, and molecular features.
• Current methods struggle to capture diverse feature interactions, limiting predictive accuracy.

Purpose of the Study:

• To develop a unified framework, PSO-FeatureFusion, for joint feature integration and optimization from multiple biological entities.
• To enhance the modeling of complex biological relationships and improve predictive performance in bioinformatics tasks.

Main Methods:

• Combines particle swarm optimization (PSO) with neural networks for feature integration and optimization.
• Models pairwise feature interactions and learns optimal contributions in a task-agnostic manner.
• Applies the framework to drug-drug interaction and drug-disease association prediction tasks.

Main Results:

• Achieved strong performance across evaluation metrics on benchmark datasets for both prediction tasks.
• Outperformed or matched state-of-the-art deep learning and graph-based models.
• Demonstrated robustness with minimal hyperparameter tuning and flexibility across varied feature structures.

Conclusions:

• PSO-FeatureFusion offers a scalable and practical solution for high-dimensional biological data analysis.
• The framework's adaptability and interpretability support applications in drug discovery and disease prediction.
• Provides an effective approach for capturing interdependencies within biological data.