TUnA: an uncertainty-aware transformer model for sequence-based protein-protein interaction prediction

Area of Science:

• Computational Biology
• Bioinformatics
• Machine Learning

Background:

• Protein-protein interactions (PPIs) are crucial for cellular functions.
• Predicting PPIs from sequence data is challenging for current deep learning models.
• Existing models struggle with generalization to unseen proteins and lack uncertainty quantification.

Purpose of the Study:

• To develop an advanced deep learning model for accurate PPI prediction.
• To enhance model generalization to novel protein sequences.
• To provide reliable uncertainty estimates for PPI predictions.

Main Methods:

• Utilized a Transformer-based architecture (TUnA).
• Incorporated ESM-2 protein embeddings and Transformer encoders.
• Integrated a Spectral-normalized Neural Gaussian Process for uncertainty estimation.

Main Results:

• Achieved state-of-the-art performance in PPI prediction.
• Successfully predicted interactions for unseen protein sequences.
• Demonstrated that uncertainty estimates effectively reduce false positives.

Conclusions:

• TUnA offers a robust solution for accurate and reliable PPI prediction.
• Uncertainty quantification is key to improving the utility of computational predictions.
• TUnA bridges the gap between computational predictions and experimental validation in PPI research.