Predicting the protein interaction landscape of a free-living bacterium with pooled-AlphaFold3

Area of Science:

• Computational biology
• Genomics
• Structural biology

Background:

• Protein-protein interactions (PPIs) are crucial for cellular functions.
• Predicting PPIs using protein complex structure prediction tools like AlphaFold3 is computationally intensive for genome-wide studies.
• Previous methods lacked scalability and efficiency for comprehensive genomic analysis.

Purpose of the Study:

• To introduce a novel, scalable method for genome-wide PPI prediction.
• To overcome the computational limitations of traditional pairwise prediction approaches.
• To generate a comprehensive PPI map for Mycoplasma genitalium.

Main Methods:

• Developed pooled-PPI prediction, a technique optimizing AlphaFold3 for large-scale PPI screening.
• Applied pooled-PPI prediction to predict all pairwise PPIs in Mycoplasma genitalium.
• Analyzed the resulting dataset for accuracy, biases, and biological insights.

Main Results:

• Pooled-PPI prediction significantly improved accuracy and reduced computational time (~twofold) and job requirements (~100-fold) compared to paired approaches.
• A comprehensive PPI map of 113,050 interactions in M. genitalium was generated using only 2027 AlphaFold3 jobs.
• The study identified a widespread size bias in AlphaFold interface scores and uncovered new biological insights in M. genitalium.

Conclusions:

• Pooled-PPI prediction is a highly scalable and efficient method for uncovering protein-protein interactions.
• This technique is a valuable addition to the functional genomics toolkit for large-scale biological discovery.
• The generated dataset provides a foundation for further research into M. genitalium biology and PPI networks.