HSEpred

HSEpred predicts half-sphere exposure (HSE) measures and residue contact numbers from protein sequences to quantify sequence–structure relationships.

Key Features:

• Two-dimensional solvent exposure (HSE-up/HSE-down): HSE divides an amino acid's surrounding sphere into upward (HSE-up) and downward (HSE-down) half-spheres to characterize local spatial neighborhoods.
• Predictive performance: Predicted-versus-observed correlation coefficients are 0.72 for HSE-up and 0.68 for HSE-down.
• Improved structural sensitivity: HSE-up and HSE-down provide enhanced performance compared to accessible surface area, residue depth, and contact number.
• Residue contact number prediction: Residue contact numbers are inferred by summing predicted HSE-up and HSE-down values.
• Machine-learning model: Uses support vector regression (SVR) to model relationships between amino acid sequences and HSE measures.

Scientific Applications:

• Sequence–structure quantification: Quantifies protein sequence–structure relationships by predicting solvent exposure profiles from sequence.
• Structural-property profiling: Predicts residue-level structural properties, including solvent exposure and contact number, from sequence alone.
• Folding, stability, and interaction analysis: Supports analysis of protein folding, stability, and residue interaction patterns.

Methodology:

Support vector regression (SVR) models were trained on a well-prepared non-homologous protein structure dataset using five different sequence-encoding schemes.