Pattern to Knowledge: Deep Knowledge-Directed Machine Learning for Residue-Residue Interaction Prediction.

Area of Science:

• Computational Biology
• Structural Biology
• Bioinformatics

Background:

• Residue-residue close contact (R2R-C) data from protein-protein interaction (PPI) experiments is used for residue-residue interaction (R2R-I) prediction.
• Complex physiochemical environments often mask R2R-I patterns in acquired data.

Purpose of the Study:

• To develop a novel method, Pattern to Knowledge (P2K), to disentangle R2R-I patterns from experimental data.
• To leverage discovered deep knowledge for sequence-based R2R-I prediction using machine learning models.

Main Methods:

• P2K disentangles R2R-I patterns to extract succinct discriminative information from different statistical/functional spaces.
• Machine learning models were constructed using this deep knowledge for sequence-based R2R-I prediction.
• A stringent leave-one-complex-out-alone cross-validation was employed on a benchmark dataset.

Main Results:

• The P2K method successfully identified and leveraged deep knowledge not explicitly present in the raw data.
• The developed R2R-I predictor demonstrated superior performance compared to existing sequence-based methods.
• The predictor achieved a 28% improvement in performance over a current sequence-based R2R-I predictor (p: 1.9E-08).

Conclusions:

• P2K offers a novel approach to uncover hidden residue-residue interaction patterns in PPI data.
• The method facilitates accurate sequence-based R2R-I prediction by utilizing extracted deep knowledge.
• P2K provides a valuable tool for advancing the understanding of protein-protein interactions.