Predicting Protein-Protein Interactions from Matrix-Based Protein Sequence Using Convolution Neural Network and...

Area of Science:

• Computational biology
• Bioinformatics
• Machine learning in life sciences

Background:

• Protein-protein interactions (PPIs) are crucial for biological processes, disease understanding, and drug development.
• Experimental identification of PPIs is limited by time, cost, and accuracy.
• Computational methods are needed to supplement experimental PPI prediction.

Purpose of the Study:

• To propose a novel computational approach, CNN-FSRF, for accurate protein-protein interaction prediction.
• To leverage deep learning and feature selection for enhanced PPI prediction from protein sequences.

Main Methods:

• Protein sequences are converted into Position-Specific Scoring Matrices (PSSM).
• A Convolutional Neural Network (CNN) is used for deep feature extraction.
• Feature-Selective Rotation Forest (FSRF) is employed to remove redundant features and improve accuracy.

Main Results:

• CNN-FSRF achieved high prediction accuracies: 97.75% on Yeast and 88.96% on Helicobacter pylori datasets.
• The method demonstrated superior performance compared to Support Vector Machines (SVM) and other existing approaches.
• Validation on independent datasets confirmed the robustness and effectiveness of CNN-FSRF.

Conclusions:

• CNN-FSRF provides a rapid and accurate computational tool for predicting protein-protein interactions.
• The proposed method serves as a valuable complement to traditional experimental techniques in biological research.
• This approach has significant implications for advancing our understanding of molecular mechanisms and accelerating drug discovery.