Protein-Protein Interaction Prediction Based on Spectral Radius and General Regression Neural Network

Area of Science:

• Computational biology
• Bioinformatics
• Biochemistry

Background:

• Protein-protein interactions (PPIs) are fundamental to cellular processes and disease mechanisms.
• Experimental PPI prediction is time-consuming and labor-intensive.
• Computational methods are crucial for efficient PPI prediction.

Purpose of the Study:

• To develop a novel computational algorithm, GRNN-PPI, for predicting protein-protein interactions.
• To utilize only amino acid sequence information for PPI prediction.
• To enhance the accuracy and efficiency of PPI prediction compared to existing methods.

Main Methods:

• Developed a new feature extraction method, Mutation Spectral Radius (MSR), using the BLOSUM62 matrix for evolutionary information.
• Integrated autocorrelation description for comprehensive physicochemical and sequence information extraction.
• Employed Principal Component Analysis for noise reduction and General Regression Neural Network (GRNN) as the classifier.

Main Results:

• Achieved high prediction accuracies: 97.47% for yeast, 99.63% for human, and 99.97% for *Helicobacter pylori*.
• Demonstrated superior performance on PPI networks and independent datasets compared to state-of-the-art methods.
• Validated the feasibility and robustness of the GRNN-PPI algorithm.

Conclusions:

• GRNN-PPI is a highly accurate and robust method for predicting protein-protein interactions using sequence data.
• The proposed MSR feature extraction method effectively captures evolutionary information.
• This computational approach offers a significant advancement for biological and disease mechanism research.