Biological network topology features predict gene dependencies in cancer cell-lines

Area of Science:

• Computational biology
• Systems biology
• Genomics

Background:

• Protein-protein interaction (PPI) networks are valuable for predicting essential proteins.
• Generic PPI networks fail to capture cell-specific genetic dependencies, especially in cancer.
• Personalized PPI networks are needed to identify context-specific dependencies.

Purpose of the Study:

• To develop a method for personalizing PPI networks using cell-specific gene expression and mutation data.
• To predict novel gene dependencies in a cell-specific manner.
• To improve the accuracy of essential protein identification for therapeutic targeting.

Main Methods:

• Personalized PPI networks were constructed by integrating cell-specific gene expression and mutation data.
• 12 topological features of PPIs were extracted.
• Random-forest classifiers (DependANT) were trained using PPI features and DepMap gene dependency data.

Main Results:

• DependANT significantly improved the prediction of common gene dependencies compared to baseline generic PPI models (up to 10.8% increase in predictive power).
• DependANT demonstrated higher sensitivity in predicting genes with dependencies in a small number of cell types.
• The study successfully predicted novel gene dependencies using personalized networks.

Conclusions:

• Personalized PPI networks offer a more accurate approach to identifying essential proteins than generic networks.
• DependANT provides a powerful tool for predicting cell-specific gene dependencies, with implications for targeted cancer therapies.
• The developed method facilitates the identification of proteins crucial for cancer cell survival.