Protein interactions in human pathogens revealed through deep learning.

Area of Science:

• Computational biology
• Structural biology
• Infectious disease research

Background:

• Understanding bacterial protein-protein interactions (PPIs) is crucial for deciphering pathogenicity mechanisms and developing novel therapeutics.
• Current methods for identifying and characterizing PPIs can be time-consuming and limited in scope.

Purpose of the Study:

• To develop a rapid, deep learning-based computational pipeline for proteome-wide identification and structural characterization of bacterial PPIs.
• To leverage residue-residue coevolution and protein structure prediction for enhanced accuracy in PPI prediction.

Main Methods:

• Development of RoseTTAFold2-Lite, a deep learning model integrating coevolutionary data and structure prediction.
• Systematic screening of 78 million protein pairs across 19 human bacterial pathogens.
• Experimental validation of selected predicted PPIs.

Main Results:

• Identification of 1,923 confidently predicted protein complexes involving essential genes.
• Discovery of 256 predicted complexes associated with virulence factors.
• Experimental validation of 6 out of 12 tested novel PPI predictions, demonstrating high accuracy.

Conclusions:

• The RoseTTAFold2-Lite pipeline offers a powerful approach for large-scale bacterial PPI discovery.
• The identified PPIs provide insights into essential cellular processes and virulence mechanisms in bacterial pathogens.
• These findings can guide the development of targeted antimicrobial therapies.