ABMDA

ABMDA predicts potential associations between microRNAs (miRNAs) and human diseases using an adaptive boosting framework to improve association prediction accuracy.

Key Features:

• Adaptive Boosting Technology: Integrates multiple weak classifiers (decision trees) into a strong classifier and assigns weights to samples to enhance predictive performance.
• Data Balancing with k-means Sampling: Uses random sampling based on k-means clustering to balance positive and negative sample sets for model training.
• Validation and Accuracy: Demonstrated performance with AUCs of 0.9170 (global leave-one-out) and 0.8220 (local leave-one-out), and a 5-fold cross-validation mean AUC of 0.9023 (SD 0.0016).
• Case Studies: Validated on colon neoplasms, hepatocellular carcinoma, and breast neoplasms, with 49–50 of the top 50 predicted miRNAs confirmed by databases and experimental literature.

Scientific Applications:

• miRNA–disease association discovery: Predicts candidate miRNAs associated with human diseases for downstream investigation.
• Complement to experimental methods: Provides an economical and efficient computational alternative to prioritize miRNA–disease hypotheses for experimental validation.
• Disease mechanism and therapy research: Facilitates insights into miRNA roles in complex diseases and supports development of targeted therapeutic strategies.

Methodology:

Random sampling based on k-means clustering to balance positive and negative samples; decision trees used as weak classifiers within an adaptive boosting framework that assigns sample weights to form a strong classifier; performance assessed by global and local leave-one-out cross-validation and 5-fold cross-validation.