Feature fusion with attributed deepwalk for protein-protein interaction prediction.

Area of Science:

• Computational biology
• Bioinformatics
• Systems biology

Background:

• Protein-protein interactions (PPIs) are fundamental to cellular functions and disease pathogenesis.
• Experimental PPI detection is resource-intensive; computational methods offer a scalable alternative.
• Existing computational methods often fail to capture the complexity of protein interactions due to limited feature integration.

Purpose of the Study:

• To develop a novel computational approach, FFADW (Feature Fusion Method with Attributed DeepWalk), for enhanced PPI prediction.
• To integrate diverse protein features, including sequence and network information, using a weighted fusion strategy.
• To improve the accuracy and robustness of computational PPI prediction.

Main Methods:

• FFADW integrates sequence similarity (Levenshtein distance) and network similarity (Gaussian kernel).
• A weighted fusion strategy with an adjustable parameter (α) combines these complementary features.
• Attributed DeepWalk learns low-dimensional protein embeddings from fused features for classification.

Main Results:

• FFADW demonstrated significant improvements in sample clustering across three benchmark datasets.
• The proposed method outperformed existing computational approaches for PPI prediction.
• The XGBoost classifier, utilizing FFADW features, achieved the best prediction performance.

Conclusions:

• The weighted fusion strategy effectively integrates heterogeneous protein data, reducing noise and redundancy.
• FFADW offers an improved and robust technique for computational prediction of protein-protein interactions.
• This approach enhances understanding of cellular processes and disease mechanisms through accurate PPI identification.