Machine Learning Approaches for Compound-Target Interaction Prediction: A Review.

Area of Science:

• Biochemistry
• Computational Biology
• Pharmacology

Background:

• Compound-target interaction (CTI) prediction is vital for drug discovery and understanding food bioactive compounds.
• Traditional experimental CTI identification is costly, time-consuming, and prone to errors.
• Machine learning (ML) offers a more efficient alternative to experimental CTI prediction.

Purpose of the Study:

• To review recent advancements in ML-based CTI prediction.
• To highlight the application of ML in identifying targets of food-derived bioactive compounds.
• To discuss challenges and future directions in ML-driven CTI prediction for functional food research.

Main Methods:

• Systematic review of ML approaches for CTI prediction.
• Outline of public databases and feature extraction for compounds (molecular fingerprints) and proteins (sequence-derived features).
• Elaboration on four ML types: supervised learning, matrix factorization, graph topology inference, and deep neural networks.

Main Results:

• ML methods provide efficient alternatives to experimental CTI identification.
• Specific focus on ML applications for food-derived bioactive compounds.
• Identification of key challenges including model interpretability and data dependency.

Conclusions:

• ML-based CTI prediction holds significant potential for advancing functional food research.
• Addressing challenges in data integration and interpretability is crucial for future progress.
• Improved ML models can accelerate the discovery and application of food-derived bioactive compounds.