MAGICAL: A multi-class classifier to predict synthetic lethal and viable interactions using protein-protein...

Area of Science:

• Computational biology
• Genetics
• Bioinformatics

Background:

• Synthetic lethality (SL) and synthetic viability (SV) are crucial genetic interactions for targeted cancer therapy.
• Current experimental methods for identifying SLs and SVs are costly and time-consuming.
• Existing computational tools primarily focus on binary classification of SL pairs, lacking discrimination between SL and SV.

Purpose of the Study:

• To develop a novel computational model for predicting both synthetic lethality and synthetic viability interactions.
• To identify key network properties that effectively discriminate between SL and SV interactions.
• To improve the accuracy and precision of genetic interaction prediction in cancer research.

Main Methods:

• Developed MAGICAL (Multi-class Approach for Genetic Interaction in Cancer via Algorithm Learning), a multi-class random forest machine learning model.
• Utilized network properties derived from protein-protein interactions as features for classification.
• Trained and validated the model on established genetic interaction datasets (CGIdb, BioGRID, SynLethDB) and external datasets (DepMap).

Main Results:

• MAGICAL achieved approximately 80% accuracy on the training dataset and demonstrated strong performance on independent datasets.
• Identified specific network properties, including shortest path, average neighbor2, average betweenness, average triangle, and adhesion, as significant discriminators between SL and SV.
• MAGICAL is the first multi-class model capable of identifying discriminatory features for both synthetic lethal and viable interactions.

Conclusions:

• MAGICAL provides a more accurate and precise method for predicting synthetic lethality and synthetic viability compared to existing binary classifiers.
• The model's ability to differentiate between SL and SV using network properties offers valuable insights into cancer genetics.
• This approach has the potential to accelerate the discovery of novel therapeutic targets in oncology.