BCrystal

BCrystal predicts protein crystallization propensity from sequence-based, structural, and physicochemical features using an XGBoost model with SHAP-based interpretability.

Key Features:

• Gradient Boosting Machine (XGBoost): BCrystal uses an optimized gradient boosting algorithm (XGBoost) to model complex relationships between protein features and crystallization propensity.
• Feature Extraction: The model integrates sequence-based features together with structural and physicochemical properties of proteins as input variables.
• Interpretability via SHAP: BCrystal applies the SHAP (SHapley Additive exPlanations) algorithm to quantify and explain feature contributions to individual predictions.
• Performance Metrics: Across three independent test sets, BCrystal reports an average accuracy of 93.7%, recall of 96.63%, and Matthew's correlation coefficient of 0.868, exceeding state-of-the-art sequence-based methods by more than 12.5% in accuracy, 18% in recall, and 0.253 in Matthew’s correlation coefficient.

Scientific Applications:

• Protein crystallization screening: Predicts which proteins are likely to crystallize to prioritize experimental trials and reduce screening effort.
• Variant prioritization for crystallizability: Screens sequence variants for enhanced crystallizability to inform construct design for structural studies.
• Structural biology target selection: Assists selection of targets for structural determination by estimating crystallization potential from sequence and physicochemical data.

Methodology:

BCrystal trains and fine-tunes an XGBoost model on a dataset comprising sequence-based, structural, and physicochemical protein features and applies SHAP to provide explanations for predictions on new proteins.