HeteroMirPred

HeteroMirPred classifies microRNA precursors (pre-miRNAs) using an ensemble of heterogeneous classifiers to distinguish pre-miRNA hairpins from non-pre-miRNA hairpin-forming sequences across human, plant, animal, and viral datasets.

Key Features:

• Ensemble Composition: Integrates Support Vector Machine (SVM), k-Nearest Neighbors (kNN), and Random Forest (RF) classifiers.
• Prediction Aggregation: Aggregates classifier outputs through a voting system to enhance classification performance.
• Training Data Scope: Trained on diverse pre-miRNA datasets including human, plant, animal, and viral sequences.
• Discriminative Features: Utilizes structural robustness features including self-containment and its derivatives to discriminate pre-miRNAs from other stem loops.
• Feature Selection: Applies correlation-based feature selection (CFS) combined with a genetic algorithm (GA) search method to identify relevant features.
• Class Rebalancing: Employs a modified Synthetic Minority Oversampling Technique (SMOTE) bagging rebalancing method to address class imbalance.
• Validation and Performance: Validated with 10 runs of 5-fold cross-validation achieving overall accuracy 96.54%, sensitivity 94.8%, specificity 98.3%, and >93% accuracy on animal, plant, and viral pre-miRNAs.
• Ensemble Advantage: Produces more reliable predictions than single classifiers according to comparative evaluation.

Scientific Applications:

• Pre-miRNA Classification: Classification of pre-miRNA sequences across human, plant, animal, and viral species.
• Pre-miRNA vs Non-pre-miRNA Discrimination: Distinguishing genuine pre-miRNA hairpins from other hairpin-forming sequences.
• Structural Robustness Analysis: Investigating intrinsic structural robustness characteristics of pre-miRNAs using selected discriminative features such as self-containment.

Methodology:

Ensemble of SVM, kNN, and RF classifiers with voting aggregation; discriminative features including self-containment and derivatives; correlation-based feature selection (CFS) combined with genetic algorithm (GA); modified SMOTE bagging rebalancing; validated by 10 runs of 5-fold cross-validation.