ANMDA

ANMDA predicts potential microRNA (miRNA)-disease associations by applying an anti-noise computational model to mitigate data noise and improve prediction accuracy.

Key Features:

• Noise Reduction: An anti-noise algorithm mitigates the effects of noise within datasets to enhance prediction reliability.
• Data Similarity Calculation: Similarities between miRNAs and diseases are calculated to construct features using data from HMDD v2.0.
• Sample Construction: Positive samples are derived from HMDD v2.0 and negative samples are obtained by k-means clustering of undetected miRNA-disease associations.
• Data Subset Creation: Multiple data subsets are generated via sampling with replacement for downstream processing.
• Machine Learning: Subsets are processed using Light Gradient Boosting Machine (LightGBM) models.
• Voting Methodology: A voting mechanism aggregates predictions across subsets to produce final miRNA-disease association scores.

Scientific Applications:

• miRNA-disease association discovery: Identification of candidate miRNAs associated with specific diseases based on predicted associations.
• Biomarker and therapeutic target prioritization: Prioritization of miRNAs as potential biomarkers or therapeutic targets for disease research.
• Functional research on miRNA roles: Support for studies of miRNA regulatory roles and disease mechanisms by providing high-confidence association predictions.

Methodology:

Calculate miRNA and disease similarities; derive positive samples from HMDD v2.0 and negative samples via k-means clustering of undetected associations; create multiple subsets by sampling with replacement; train and predict with LightGBM on subsets; aggregate results by voting; validated by AUROC = 0.9373 ± 0.0005 in five-fold cross-validation.