Prediction of Synthetic Lethal Interactions in Human Cancers Using Multi-View Graph Auto-Encoder

Area of Science:

• Computational biology
• Bioinformatics
• Genomics

Background:

• Synthetic lethality (SL) is key for targeted cancer therapies, but experimental detection is costly and inconsistent.
• Computational methods are emerging to predict SL interactions, complementing experimental approaches.
• Graph-based machine learning offers a promising avenue for SL prediction due to the graph structure of SL data.

Purpose of the Study:

• To develop a novel computational method for predicting synthetic lethality interactions.
• To leverage multi-view graph autoencoders and attention mechanisms for enhanced SL prediction accuracy.
• To address the limitations of current experimental and computational SL detection methods.

Main Methods:

• Proposed a Multi-view Graph Auto-Encoder (SLMGAE) model for synthetic lethality prediction.
• Utilized the SL interaction graph as the main view and incorporated other biological network graphs (e.g., PPI, GO) as support views.
• Implemented an attention mechanism to dynamically weight the importance of different support views.

Main Results:

• The SLMGAE model demonstrated superior performance compared to state-of-the-art methods on the SynLethDB dataset.
• Experimental results validated the effectiveness of SLMGAE in predicting novel synthetic lethality interactions.
• Case studies confirmed the practical utility and accuracy of the proposed SLMGAE method.

Conclusions:

• SLMGAE provides an effective and accurate computational approach for predicting synthetic lethality interactions.
• The multi-view graph autoencoder framework with attention mechanism enhances the prediction of SL relationships.
• This method holds potential for accelerating the discovery of novel anti-cancer drug targets through synthetic lethality.