Bastion3: a two-layer ensemble predictor of type III secreted effectors

Area of Science:

• Microbiology and Molecular Biology
• Bioinformatics and Computational Biology

Background:

• Type III secreted effectors (T3SEs) are crucial for Gram-negative bacterial pathogens to interact with host cells.
• Existing computational tools for T3SE identification have limitations, including reliance on partial protein sequences and limited feature sets.

Purpose of the Study:

• To develop an improved computational tool for accurate identification of T3SEs.
• To address limitations of current methods by utilizing comprehensive protein sequence features and advanced machine learning.

Main Methods:

• Developed Bastion3, a two-layer ensemble predictor utilizing a wide range of protein features.
• Employed LightGBM, a gradient boosting machine, and a genetic algorithm for parameter optimization.
• Integrated multiple single models through ensemble learning for enhanced prediction accuracy.

Main Results:

• Bastion3 achieved high performance metrics: ACC 0.959, F-value 0.958, MCC 0.917, and AUC 0.956.
• Outperformed existing methods by significant margins (e.g., >5.6% ACC).
• Developed a user-friendly online toolkit for convenient T3SE prediction.

Conclusions:

• Bastion3 represents a state-of-the-art toolkit for T3SE prediction.
• The ensemble approach and comprehensive feature set significantly enhance prediction accuracy.
• Bastion3 facilitates future discoveries of novel T3SEs and aids in understanding pathogen-host interactions.