PeNGaRoo

PeNGaRoo predicts non-classical secreted proteins in Gram-positive bacteria using a two-layer LightGBM ensemble to support studies of protein secretion mechanisms and bacterial pathogenesis.

Key Features:

• High-Quality Dataset Construction: Built from a curated dataset of experimentally verified non-classical secreted proteins and used to create benchmark datasets for training and validation.
• Advanced Feature Engineering: Performs comprehensive feature analysis and engineering to assess individual feature performance and improve predictive accuracy beyond traditional methods.
• Two-Layer LightGBM Ensemble Model: Implements a two-layer ensemble based on LightGBM that integrates multiple single-feature-based models and optimizes parameters via particle swarm optimization.
• Predictive Performance: Achieves accuracy of 0.900, F-value of 0.903, Matthew's correlation coefficient of 0.803, and area under the curve of 0.963, surpassing previous state-of-the-art predictors.

Scientific Applications:

• Discovery of Non-Classical Secreted Proteins: Enables identification of non-classical secreted proteins in Gram-positive bacteria to inform studies of bacterial pathogenesis and therapeutic target discovery.
• Protein Secretion Mechanism Studies and Biotechnology: Supports exploration of protein secretion mechanisms and applications in biotechnology through high predictive accuracy.

Methodology:

Constructs benchmark datasets from experimentally verified non-classical secreted proteins, performs feature analysis and engineering, builds a two-layer ensemble of single-feature-based LightGBM models, and optimizes ensemble parameters using particle swarm optimization.