A CSGNN model-based method for essential protein identification.

Area of Science:

• Computational Biology
• Systems Biology
• Bioinformatics

Background:

• Identifying essential proteins is crucial for understanding cellular functions and diseases.
• Existing computational methods often fail to capture dynamic expression and network context, limiting prediction accuracy.

Purpose of the Study:

• To propose a novel computational method, Correlation-guided Subgraph Graph Neural Network (CSGNN), for accurate essential protein identification.
• To improve the modeling of dynamic expression activity and global network context in essential protein prediction.

Main Methods:

• Constructed a weighted protein network using Pearson correlation coefficients from activity-aware expression matrices.
• Utilized a two-layer attention-based graph convolution to learn protein embeddings within multi-scale subgraph contexts.
• Integrated protein embeddings with neighborhood context for final essentiality probability prediction using a multilayer perceptron.

Main Results:

• CSGNN effectively integrates correlation-guided graph construction with attention-based representation learning.
• The method accurately captures dynamic expression patterns and global network information.
• Experiments on yeast and E. coli datasets showed CSGNN outperformed traditional methods in essential protein identification.

Conclusions:

• CSGNN offers a robust and accurate approach for identifying essential proteins.
• The method enhances understanding of cellular processes and disease mechanisms by improving essential protein prediction.
• CSGNN demonstrates superior performance and robustness compared to existing computational baselines.