An Efficient Feature Extraction Technique Based on Local Coding PSSM and Multifeatures Fusion for Predicting...

Area of Science:

• Biomedical informatics
• Computational biology
• Proteomics

Background:

• Protein-protein interactions (PPIs) are crucial for cellular structure and function.
• Accurate prediction of PPIs is vital for biomedical research.
• Existing machine learning methods for PPI prediction have limitations.

Purpose of the Study:

• To develop a novel sequence-based feature extraction method for enhanced PPI prediction.
• To improve the accuracy and efficiency of identifying protein-protein interactions.
• To provide a valuable tool for future proteomics research.

Main Methods:

• Proposed LCPSSMMF: a method combining local coding position-specific scoring matrix (PSSM) with multi-feature fusion.
• Developed a new PSSM (CPSSM) using a local coding method integrating global and local feature extraction.
• Employed Local Average Group (LAG) and Bigram Probability (BP) for feature extraction from CPSSM.
• Utilized multi-feature fusion to create final feature vectors.

Main Results:

• Achieved prediction accuracies of 93.43% for yeast and 90.41% for human PPI datasets.
• Demonstrated superior performance compared to existing state-of-the-art sequence-based methods on yeast datasets.
• The LCPSSMMF approach effectively captures both local and global discriminatory information for PPI identification.

Conclusions:

• The LCPSSMMF method offers a significant advancement in predicting protein-protein interactions.
• The developed LCPSSMMFSVM server is freely available to facilitate further research in proteomics.
• This approach enhances the understanding of cellular organization through accurate PPI prediction.