TXSelect: A multi-task learning model to identify secretory effectors.

Area of Science:

• Microbiology
• Bioinformatics
• Computational Biology

Background:

• Pathogenic microorganisms utilize secretory effectors to manipulate host processes, influencing survival and pathogenicity.
• Accurate classification of diverse bacterial effectors (types I, II, III, IV, and VI secretory effectors - TXSE) is challenging due to sequence and structural heterogeneity.

Purpose of the Study:

• To develop an efficient computational framework, TXSelect, for the simultaneous classification of multiple TXSE types.
• To integrate advanced protein features for improved classification accuracy and biological insight.

Main Methods:

• Developed TXSelect, a multi-task learning framework with a shared backbone and task-specific heads for TXSE classification.
• Integrated protein embedding features from Evolutionary Scale Modelling (ESM) N-terminal mean with classical descriptors: Distance-based Residue (DR) and Split Amino Acid Composition General (SC-PseAAC-General).
• Evaluated feature combinations and model performance using rigorous validation and testing, including Uniform Manifold Approximation and Projection for visualization.

Main Results:

• The optimal feature combination (ESM N-terminal mean + DR + SC-PseAAC) achieved high accuracy, with validation F1 score of 0.867 and test F1 score of 0.8645.
• TXSelect demonstrated robust generalization capabilities across different TXSE types.
• Model interpretability and discriminative power were validated through comprehensive assessments and visualization techniques.

Conclusions:

• TXSelect provides an accurate and efficient computational tool for classifying bacterial TXSE.
• The framework supports deeper biological understanding of pathogen-host interactions.
• This tool has potential applications in identifying therapeutic targets for infectious diseases.