PBHMDA: Path-Based Human Microbe-Disease Association Prediction.

Area of Science:

• Microbiology and Computational Biology
• Human Health and Disease Mechanisms

Background:

• Microorganisms are increasingly linked to human diseases, offering insights into pathogenesis, diagnosis, and precision medicine.
• Current knowledge of microbe-disease associations is limited, necessitating advanced computational approaches for discovery.

Purpose of the Study:

• To develop and validate a computational model, Path-Based Human Microbe-Disease Association prediction (PBHMDA), for predicting microbe-disease associations.
• To leverage known associations and network analysis to identify novel microbial contributors to human diseases.

Main Methods:

• Integrated known microbe-disease associations with Gaussian interaction profile kernel similarity for microbes and diseases.
• Employed a depth-first search algorithm to traverse paths and infer disease-related microbes.
• Validated performance using global, local leave-one-out, and 5-fold cross-validation, calculating Area Under the ROC Curve (AUC).

Main Results:

• PBHMDA achieved high prediction accuracy with AUCs of 0.9169 (global LOOCV) and 0.8767 (local LOOCV).
• 5-fold cross-validation yielded an average AUC of 0.9082 ± 0.0061, confirming model efficiency.
• Case studies for liver cirrhosis, type 1 diabetes, and asthma showed high validation rates for top-ranked microbe predictions (9/10, 7/10, 9/10).

Conclusions:

• PBHMDA demonstrates significant potential for discovering novel microbe-disease associations.
• The model aids in prioritizing potential microbe-disease pairs for experimental validation, advancing understanding of microbial roles in human diseases.
• The PBHMDA tool and data are publicly available to support further research.