PPIscreenML is a method for structure-based screening of protein-protein interactions using AlphaFold.

Area of Science:

• Biochemistry
• Structural Biology
• Bioinformatics

Background:

• Protein-protein interactions are fundamental to cellular functions.
• Accurate protein complex modeling is now possible with tools like AlphaFold2 (AF2).
• Distinguishing true interactions from decoys in AF2 models remains a challenge.

Purpose of the Study:

• To develop and benchmark a method for inferring protein-protein interactions from AF2 models.
• To create a classification model that distinguishes interacting pairs from non-interacting decoys.
• To enable screening of candidate protein pairings for building interaction networks.

Main Methods:

• Trained a classification model, PPIscreenML, using AF2 models of known interacting and decoy protein pairs.
• Evaluated PPIscreenML performance against existing methods like pDockQ and iPTM.
• Tested generalization on structurally conserved protein superfamilies, including the tumor necrosis factor superfamily (TNFSF).

Main Results:

• PPIscreenML significantly outperforms pDockQ and iPTM in identifying true protein-protein interactions.
• The model demonstrates high accuracy in predicting ligand/receptor interactions within the TNFSF.
• Performance on unseen complexes indicates strong generalization capabilities.

Conclusions:

• PPIscreenML provides a robust, benchmarked approach for identifying protein-protein interactions using structural models.
• The tool is broadly applicable for discovering novel protein complexes predicted by AF2.
• This advances the use of structural modeling for mapping cellular interaction networks.