Mining protein networks for synthetic genetic interactions

Area of Science:

• Systems Biology
• Computational Biology
• Genomics

Background:

• Protein interaction networks reveal functional properties, including essentiality.
• Understanding joint dispensability of protein pairs is crucial for synthetic sick/lethal interactions.
• Saccharomyces cerevisiae offers a model for studying synthetic genetic interactions due to available data.

Purpose of the Study:

• To assess the predictive power of protein interaction networks for synthetic genetic interactions.
• To determine if network properties of two proteins predict their joint dispensability.
• To identify novel synthetic sick/lethal gene pairs using computational methods.

Main Methods:

• Developed a support vector machine system using graph-theoretic properties from protein interaction networks.
• Trained the system on interacting and non-interacting gene pairs from genetic screens and literature data.
• Evaluated prediction performance using sensitivity and specificity metrics.

Main Results:

• The prediction system achieved over 85% sensitivity and specificity for synthetic genetic interactions.
• Prediction performance demonstrated robustness against errors in protein interaction networks and dataset variations.
• Genome-wide predictions identified novel synthetic sick/lethal gene pairs with properties similar to known pairs.

Conclusions:

• Protein interaction networks effectively predict synthetic lethal interactions, matching or exceeding other computational methods.
• Network-based predictions offer valuable insights into epistatic effects among genes.
• Protein interaction networks are rich sources of information for understanding genetic interactions.