ISPRED4: interaction sites PREDiction in protein structures with a refining grammar model

Area of Science:

• Computational biology
• Structural bioinformatics
• Machine learning in bioinformatics

Background:

• Identifying protein-protein interaction (PPI) sites is crucial for understanding cellular functions.
• Experimental methods for PPI site identification are resource-intensive.
• Computational tools offer an efficient alternative for predicting PPI sites.

Purpose of the Study:

• To present ISPRED4, an improved predictor of PPI sites.
• To enhance the prediction of protein functional integration using computational approaches.

Main Methods:

• Developed ISPRED4, a structure-based predictor for PPI sites on unbound monomer surfaces.
• Employed machine learning algorithms incorporating features from protein sequence and structure.
• Validated performance using cross-validation on a dataset of 151 high-resolution protein complexes.

Main Results:

• ISPRED4 achieved a per-residue Matthew Correlation Coefficient of 0.48 and an overall accuracy of 0.85.
• Benchmarking demonstrated ISPRED4 as a top-performing PPI site predictor.
• The predictor effectively identifies PPI sites on unbound monomer surfaces.

Conclusions:

• ISPRED4 represents a significant advancement in computational prediction of PPI sites.
• The tool aids in characterizing protein functional integration.
• ISPRED4 is publicly available for research use.