Human Microbe-Disease Association Database (HMDAD)

Human Microbe-Disease Association Database (HMDAD) integrates curated microbe–disease association data and implements the NCPHMDA (Network Consistency Projection for Human Microbe-Disease Association Prediction) algorithm to predict potential associations between microbes and human diseases for large-scale analysis.

Key Features:

• NCPHMDA algorithm: Uses network-consistency projection to predict microbe–disease associations.
• Gaussian interaction profile kernel similarity: Integrates Gaussian interaction profile kernel similarity for both microbes and diseases with known associations.
• Biological assumption: Operates on the assumption that microbes with similar functions exhibit analogous association or non-association patterns with similar diseases, and vice versa.
• Non-parametric universal network-based prediction: Functions as a non-parametric universal network-based method capable of predicting associated microbes for multiple diseases simultaneously without requiring negative samples.
• Cross-validation performance: Reported AUCs of 0.9039 (global leave-one-out CV), 0.7953 (local leave-one-out CV), and an average AUC of 0.8918 (5-fold CV).
• Case-study validation: Independent validations showed literature confirmation rates of nine out of ten top predictions for colon cancer, nine out of ten for asthma, and eight out of ten for type 2 diabetes.

Scientific Applications:

• Large-scale association prediction: Prioritizes candidate microbe–disease associations across many diseases for downstream analysis.
• Candidate prioritization for specific diseases: Identifies and ranks microbial candidates relevant to colon cancer, asthma, and type 2 diabetes based on predicted association scores.
• Support for experimental design: Provides prioritized hypotheses to guide clinical and experimental validation studies of microbe–disease relationships.

Methodology:

NCPHMDA integrates known microbe–disease associations with Gaussian interaction profile kernel similarity for microbes and diseases and projects potential associations in a network-consistent manner as a non-parametric universal network-based method, evaluated using global leave-one-out cross-validation, local leave-one-out cross-validation, and 5-fold cross-validation.