NAGbinder

NAGbinder predicts N-acetylglucosamine (NAG) interacting residues in protein sequences to identify potential NAG-binding sites for studies of glycosylation, protein function, and disease mechanisms.

Key Features:

• Data-Driven Approach: Derived from 231 nonredundant NAG-interacting protein chains extracted from the Protein Data Bank (PDB) with a maximum of 40% sequence identity, forming the basis for training, validation, and evaluation.
• Model Development: Trained on a balanced dataset of 1,335 NAG-interacting and 1,335 noninteracting residues, with various window sizes tested, using Random Forest models built on binary residue profiles.
• Performance Metrics: Achieved Matthews Correlation Coefficient (MCC) of 0.31 (training) and 0.25 (validation) with Area Under Receiver Operating Curve (AUROC) of 0.73 (training) and 0.70 (validation); on a realistic dataset of 1,335 interacting versus 47,198 noninteracting residues, reported MCCs of 0.26 (training) and 0.27 (validation) with AUROCs of 0.70 (training) and 0.71 (validation).
• Practical Prediction Rate: When applied to a 1,000 amino acid sequence, approximately five out of ten predicted NAG-interacting residues are correctly identified.

Scientific Applications:

• Glycosylation studies: Identifies putative NAG-binding sites to inform analyses of protein glycosylation.
• Protein structure and function: Aids structural characterization and functional inference by highlighting likely NAG-interacting residues.
• Disease and therapeutic research: Supports investigation of disease mechanisms involving NAG interactions and the development of therapeutics targeting glycoproteins.

Methodology:

Extraction of 231 nonredundant NAG-interacting chains from the PDB with ≤40% sequence identity; creation of a balanced dataset of 1,335 interacting and 1,335 noninteracting residues; testing of various window sizes; Random Forest model training using binary residue profiles; evaluation using MCC and AUROC including assessment on a realistic 1,335 versus 47,198 residue dataset.