NonClasGP-Pred

NonClasGP-Pred predicts non-classically secreted proteins (NCSPs) in Gram-positive bacteria to identify extracellular proteins lacking signal peptides and support analysis of their roles in virulence and cell defense.

Key Features:

• Balanced Subdataset Generation: Constructs ten balanced subdatasets to reduce prediction bias during ensemble model training.
• Enhanced Feature Representation: Applies the F-score algorithm with sequential forward search for feature selection across subdatasets.
• Subset-Specific Optimal Feature Combination: Integrates subset-specific optimal models to analyze imbalanced datasets from multiple perspectives.
• Unified Model Integration: Combines subdataset-based models into a unified ensemble model named NonClasGP-Pred.
• Validation Metrics: Reports accuracy 93.23%, sensitivity 100%, specificity 89.01%, Matthew's correlation coefficient 87.68%, and AUC 0.9975 based on ten-fold cross-validation.

Scientific Applications:

• NCSP identification in Gram-positive bacteria: Enables detection of extracellular proteins without conventional signal peptides to investigate their biological functions.
• Study of bacterial virulence and cell defense: Supports analysis of proteins implicated in pathogenicity and host–microbe interactions.
• Bacterial protein localization and function analyses: Provides predictions useful for downstream experimental design and functional annotation.

Methodology:

Constructs ten balanced subdatasets; uses the F-score algorithm with sequential forward search for feature selection; develops subset-specific optimal models and integrates them into a unified ensemble (NonClasGP-Pred); evaluates performance by ten-fold cross-validation.