Combining structural modeling and deep learning to calculate the E. coli protein interactome and functional networks

Area of Science:

• Computational biology
• Bioinformatics
• Systems biology

Background:

• Understanding protein-protein interactions (PPIs) is crucial for deciphering cellular mechanisms.
• Existing prediction methods have limitations in accuracy and scope.
• A proteome-wide approach is needed to map complex biological networks.

Purpose of the Study:

• To develop and validate an integrated computational method for predicting binary protein-protein complexes.
• To enhance the accuracy and confidence of protein-protein interaction predictions.
• To explore functional subnetworks within the E. coli interactome using predicted interactions.

Main Methods:

• Integration of three distinct prediction algorithms: PrePPI (3D structure), Topsy-Turvy (protein language model), and ZEPPI (evolutionary information).
• Validation using the high-quality HINT database of known binary PPIs.
• Application of the AF3Complex algorithm for PPI structure prediction and interactome clustering.

Main Results:

• The integrated method significantly outperforms individual component methods in predicting binary PPIs.
• A higher number of high-confidence protein-protein interactions were identified compared to single methods.
• Prediction of 374 PPI structures, with significant overlap between AF3Complex and PrePPI models.
• Clustering of the E. coli interactome identified 385 functionally coherent subnetworks.

Conclusions:

• The integrated approach offers a robust and accurate method for large-scale PPI prediction.
• The identified subnetworks provide valuable insights into cellular organization and function.
• This work facilitates the annotation of proteins with unknown functions and advances systems biology research.