Predicting host-pathogen interactions with machine learning algorithms: A scoping review.

Area of Science:

• Microbiology and Immunology
• Computational Biology
• Artificial Intelligence in Medicine

Background:

• Pathogenic microorganisms present a global health challenge, necessitating understanding of host-pathogen interactions (HPIs).
• Protein-protein interactions (PPIs) are key to HPIs, crucial for therapeutic development.
• Experimental methods for HPIs are labor-intensive; AI and machine learning offer efficient prediction.

Purpose of the Study:

• To systematically review and evaluate machine learning methodologies for Host-Pathogen Interaction (HPI) prediction.
• To categorize existing studies by host/pathogen types, algorithms, and evaluation metrics.
• To identify challenges and provide a roadmap for future research in AI-driven HPI prediction.

Main Methods:

• Scoping review of machine learning-based HPI prediction studies from 2019-2024.
• Searched reputable databases using keywords related to HPIs.
• Selected 30 out of 46 relevant articles based on title and abstract evaluation.

Main Results:

• Tree-based algorithms (Random Forest, Gradient Boosting) are most prevalent in HPI prediction.
• Deep learning models (CNNs, RNNs) show promise but require substantial labeled data.
• Significant gaps exist in dataset standardization and model interpretability.

Conclusions:

• Machine learning holds significant potential for HPI prediction.
• Addressing challenges in dataset quality, feature selection, and model transparency is crucial.
• This review offers a systematic comparison of computational approaches, guiding future AI-driven pathogen research.