ProteinPrompt: a webserver for predicting protein-protein interactions

Area of Science:

• Bioinformatics
• Computational Biology
• Drug Discovery

Background:

• Protein-protein interactions (PPIs) are crucial for cellular functions.
• Understanding PPIs is vital for designing therapeutics and identifying drug side effects.
• Identifying PPIs is a complex and time-consuming process.

Purpose of the Study:

• To develop a webserver, ProteinPrompt, for predicting unknown protein-protein interactions.
• To accelerate the identification of drug targets and potential binding partners.
• To provide a reliable tool for scanning large sequence libraries.

Main Methods:

• Utilized machine learning algorithms, including random forest (RF) and graph neural network (GNN).
• Developed a comprehensive, filtered database of known protein binders.
• Implemented two complementary search methods (RF and GNN) with a consensus prediction option.

Main Results:

• The RF method achieved an accuracy of 0.88 and an AUC of 0.95 on a challenging test dataset.
• The GNN method achieved an accuracy of 0.86.
• The consensus prediction reached an accuracy of 0.89, reducing the likelihood of errors.

Conclusions:

• ProteinPrompt offers a fast, accurate, and user-friendly online service for predicting PPIs.
• The webserver can scan the human proteome for potential binding partners within minutes.
• A Docker image is available for local, offline batch submission.