PPIGAN: Prediction of Protein-Protein Interactions Using Generative Adversarial Networks

Area of Science:

• Computational Biology
• Bioinformatics
• Machine Learning in Biology

Background:

• Protein-protein interactions (PPIs) are crucial for understanding cellular mechanisms.
• Accurate negative datasets are essential for evaluating PPI prediction models.
• Current random sampling methods for negative dataset construction suffer from unstable prediction accuracy.

Purpose of the Study:

• To address the limitations of random sampling in constructing negative datasets for PPI prediction.
• To propose a novel method, PPIGAN, for generating high-quality negative samples.
• To improve the accuracy and generalization ability of PPI prediction models.

Main Methods:

• Developed PPIGAN, a method based on conditional generative adversarial networks (CGANs).
• Utilized a generative network to create negative samples for PPI prediction.
• Employed a competitive learning process between the generator and the PPI prediction model.

Main Results:

• Achieved high prediction accuracy: 94.68% on yeast datasets and 98.22% on human datasets via 5-fold cross-validation.
• Demonstrated superior or comparable performance against advanced models like PIPR, CNN, DeepTrio, and DeepFE.
• Showcased enhanced model generalization ability and prediction accuracy through adversarial training.

Conclusions:

• PPIGAN offers an effective solution for constructing negative datasets in PPI prediction.
• The proposed method significantly improves the accuracy and reliability of PPI prediction.
• This work provides a valuable tool for researchers investigating molecular mechanisms through PPI analysis.