BGSVM-NUC

BGSVM-NUC predicts protein-nucleotide binding sites to identify binding residues for protein function analysis and drug design.

Key Features:

• Prediction target: Predicts protein-nucleotide binding sites at the residue level.
• Class imbalance handling: Addresses class imbalance between nonbinding (majority) and binding (minority) residues that affects machine-learning predictors for protein-ligand binding sites.
• Algorithm: Implements Boosting Multiple Granular Support Vector Machines (BGSVM).
• Sampling strategy: Trains multiple base SVMs each on a granular subset that includes all minority samples and a selected portion of majority samples.
• Validation: Uses cross-validation and independent testing across five protein-nucleotide interaction types.
• Comparative performance: Demonstrated improved prediction performance relative to sequence-based predictors for protein-nucleotide binding sites.

Scientific Applications:

• Protein function elucidation: Identifies nucleotide-binding residues to support annotation of protein functions.
• Drug design: Provides residue-level binding predictions to inform drug-target interaction studies involving nucleotides.
• Imbalance learning in bioinformatics: Serves as an approach for mitigating class imbalance in binding-site prediction datasets.

Methodology:

Uses Boosting Multiple Granular Support Vector Machines by training multiple base SVMs on granular subsets that include all minority samples and selected majority samples; validation performed via cross-validation and independent testing across five protein-nucleotide interaction types.