NetSurfP

NetSurfP predicts the relative surface accessibility of amino acids in proteins and provides a per-residue Z-score reliability measure to quantify prediction confidence.

Key Features:

• Predictive Accuracy: NetSurfP uses an ensemble of artificial neural networks trained on experimentally solved protein structures to estimate per-residue relative surface accessibility.
• Reliability Scoring: NetSurfP assigns a Z-score to each prediction during training to quantify the confidence of per-residue accessibility estimates.
• Performance Evaluation: NetSurfP was evaluated on the CB513 benchmark set and achieved a Pearson's correlation coefficient of 0.72, comparable to Real‑SPINE.
• Enhanced Discrimination: Sorting predictions by Z-score improves discrimination between high- and low-quality predictions, yielding a Pearson's correlation of 0.79 for the top 20% of predictions versus 0.74 for Real‑SPINE.

Scientific Applications:

• Protein-Protein Interaction Studies: Predicts surface-exposed amino acids to support identification of interaction sites and binding interfaces.
• Drug Design and Development: Identifies surface-accessible regions that may serve as potential small-molecule binding sites.
• Structural Biology: Provides residue exposure information to aid interpretation of X-ray crystallography and cryo-electron microscopy data and studies of protein folding and stability.

Methodology:

NetSurfP employs an ensemble of artificial neural networks trained on experimentally solved protein structures, assigns a per-residue Z-score reliability metric during training, and was evaluated using the CB513 benchmark set.