Harnessing Terminal Signal-Aware Deep Learning for Accurate Multi-Class Secreted Effector Identification

Area of Science:

• Microbiology
• Bioinformatics
• Computational Biology

Background:

• Gram-negative bacterial secreted effectors are vital for pathogenesis, requiring accurate identification.
• Current deep learning methods often overlook terminal regions containing critical secretion signals.
• Class imbalance among effector types poses a significant challenge for prediction models.

Purpose of the Study:

• To develop a novel framework, TermSE, for accurate multi-class identification of bacterial secreted effectors.
• To incorporate terminal sequence features alongside global representations for enhanced characterization.
• To address the challenge of class imbalance in secreted effector prediction.

Main Methods:

• TermSE utilizes convolutional neural networks on protein language model embeddings to capture N-terminal and C-terminal features.
• Global sequence representations are integrated for multi-view sequence characterization.
• A cosine-normalized classifier with weighted sampling is employed to handle class imbalance.

Main Results:

• TermSE significantly outperforms existing methods in both cross-validation and independent testing.
• The framework demonstrates robust generalization across varying protein sequence identities.
• Interpretability analysis confirms TermSE focuses on biologically relevant terminal patterns.

Conclusions:

• TermSE offers an effective and interpretable solution for secreted effector discovery.
• The terminal signal-aware approach advances the field of bacterial pathogenesis research.
• This method holds promise for identifying novel secreted effectors and understanding their functions.