DisLocate

DisLocate predicts cysteine bonding states and disulfide bond connectivity in proteins using machine-learning models informed by subcellular localization and correlated mutation analysis.

Key Features:

• Machine Learning-Based Prediction: Employs a two-step machine learning approach to predict both bonding states and connectivity patterns of cysteine residues.
• Incorporation of Subcellular Localization: Incorporates protein subcellular localization as an explicit feature, improving bonding-state accuracy by 3 percentage points, connectivity accuracy by 2 percentage points, and yielding an overall performance improvement exceeding 10 percentage points versus sequence-only methods.
• Integration with Correlated Mutation Information: Integrates correlated mutation data using corrected mutual information and the inverse of the covariance matrix to enhance disulfide topology predictions in eukaryotic proteins.
• Support Vector Regression (SVR) Enhancement: Incorporates Support Vector Regression to raise connectivity prediction accuracy from 54% to 59% (a 5 percentage-point increase).
• Improved Per-Protein Accuracy: Combining correlated mutation information achieves a per-protein accuracy of 38%, 2 percentage points higher than previously reported methods for eukaryotic disulfide location and topology prediction.

Scientific Applications:

• Protein structure–function analysis: Enables investigation of the roles of disulfide bonds in protein folding, stability, and molecular function.
• Subcellular environment studies: Facilitates assessment of how oxidizing conditions across subcellular compartments influence disulfide bond formation and stability.
• Eukaryotic disulfide topology prediction: Provides improved predictions of disulfide bond location and connectivity specifically for eukaryotic proteins.

Methodology:

Two-step machine learning prediction with explicit incorporation of subcellular localization, integration of correlated mutation information via corrected mutual information and inverse covariance matrix, Support Vector Regression for connectivity prediction, and use of GRHCRF code in development.