Identifying virulence factors using graph transformer autoencoder with ESMFold-predicted structures.

Area of Science:

• Computational biology
• Bacterial pathogenesis
• Drug discovery

Background:

• Antibiotic resistance necessitates novel therapeutic strategies.
• Antivirulence therapies targeting bacterial virulence factors (VFs) offer a promising alternative.
• Current computational VF identification methods often overlook crucial structural information.

Purpose of the Study:

• To develop a novel computational model for accurate identification and prediction of bacterial virulence factors (VFs).
• To leverage 3D protein structural data for enhanced VF prediction, moving beyond sequence-based approaches.

Main Methods:

• Proposed a graph transformer autoencoder for VF identification (GTAE-VF).
• Utilized ESMFold-predicted 3D protein structures.
• Integrated graph convolutional networks and transformers for all-pair message passing to capture complex relationships.
• Framed VF identification as a graph-level prediction task.

Main Results:

• GTAE-VF achieved a high prediction accuracy with an AUC of 0.963 on an independent test dataset.
• The model demonstrated superior performance compared to existing structure-based and sequence-based methods.
• GTAE-VF effectively learns both local and global structural information.

Conclusions:

• GTAE-VF presents a robust and accurate method for identifying bacterial virulence factors using 3D structural data.
• This approach holds significant potential for advancing antivirulence strategies and combating antibiotic resistance.
• GTAE-VF can serve as a valuable tool for researchers in assessing VFs and developing new therapeutic interventions.