GBDT-LR

GBDT-LR predicts potential microRNA (miRNA)-disease associations by combining gradient boosting decision trees for feature extraction with logistic regression for association scoring to prioritize candidate miRNAs.

Key Features:

• Hybrid Model Architecture: GBDT-LR integrates gradient boosting decision trees (GBDT) with logistic regression (LR) to combine nonlinear feature learning with parametric association scoring.
• Balanced Sample Selection: Uses k-means clustering to balance positive and negative samples by selecting negative samples from unknown miRNA-disease associations.
• Feature Extraction and Prediction: Applies GBDT to extract discriminative features from complex datasets and inputs those features into LR to compute final association scores.
• High Predictive Performance: Experimental validation using 5-fold cross-validation reports an average Area Under the Curve (AUC) of 0.9274.

Scientific Applications:

• Colon cancer case study: In a colon cancer case study, 90% of the top-ranked miRNAs were validated by existing databases.
• Gastric cancer case study: In a gastric cancer case study, 94% of the top-ranked miRNAs were validated by existing databases.
• Pancreatic cancer case study: In a pancreatic cancer case study, 88% of the top-ranked miRNAs were validated by existing databases.

Methodology:

Uses k-means clustering to select negative samples from unknown miRNA-disease pairs, applies GBDT for feature extraction, uses logistic regression to compute association scores, and evaluates performance with 5-fold cross-validation.