TT3D: Leveraging precomputed protein 3D sequence models to predict protein-protein interactions.

Area of Science:

• Computational biology
• Structural bioinformatics
• Machine learning in genomics

Background:

• High-quality computational protein structural models are widely available.
• Predicting protein-protein interactions (PPIs) efficiently using these models remains a challenge.
• The Foldseek method encodes protein structure into a 3Di sequence.

Purpose of the Study:

• To develop a method for high-throughput prediction of protein-protein interactions.
• To leverage both amino acid sequences and structural information for improved prediction accuracy.
• To enable genome-wide PPI prediction using readily available structural models.

Main Methods:

• Utilized the Topsy-Turvy deep-learning framework.
• Integrated amino acid sequences with Foldseek-generated 3Di structural sequences as input.
• Developed the TT3D (Topsy-Turvy 3D) prediction model.

Main Results:

• TT3D substantially improves cross-species protein-protein interaction prediction performance.
• The method effectively reuses precomputed structural models.
• TT3D is computationally lightweight for genome-wide PPI predictions.

Conclusions:

• TT3D offers an efficient and accurate approach for predicting protein-protein interactions.
• The method facilitates the large-scale application of structural information in understanding protein function.
• This work enables comprehensive genome-wide PPI mapping.