ASPIRER: a new computational approach for identifying non-classical secreted proteins based on deep learning

Area of Science:

• Microbiology
• Molecular Biology
• Bioinformatics

Background:

• Protein secretion is crucial for bacterial intercellular communication and environmental interaction.
• Gram-positive bacteria utilize diverse secretion pathways, with the non-classical pathway gaining research interest.
• Non-classical secreted proteins (NCSPs) lack signal peptides, making their identification challenging.

Purpose of the Study:

• To develop an improved computational tool for predicting non-classical secreted proteins (NCSPs).
• To address limitations of existing NCSP predictors that rely solely on whole amino acid sequences.
• To leverage deep learning for enhanced identification of proteins secreted via non-classical pathways.

Main Methods:

• Proposed a hybrid deep learning framework named ASPIRER.
• Combined a whole sequence-based XGBoost model with an N-terminal sequence-based convolutional neural network (CNN).
• Validated performance using 5-fold cross-validation and independent tests.

Main Results:

• ASPIRER demonstrated superior performance compared to existing state-of-the-art NCSP prediction methods.
• The hybrid approach effectively utilizes both whole and N-terminal sequence information for prediction.
• Achieved high accuracy in identifying putative NCSPs.

Conclusions:

• ASPIRER offers a significant advancement in predicting non-classical secreted proteins (NCSPs).
• The tool facilitates the discovery of novel NCSPs and aids in prioritizing candidates for experimental validation.
• Publicly available code and datasets promote further research in bacterial protein secretion.