MGANSL: multi-network representation generating with generative adversarial network for synthetic lethality...

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Cancer arises from multiple mutations, necessitating targeted therapies like synthetic lethality (SL).
• SL involves targeting cancer driver genes with their synthetic lethal partners.
• Existing computational methods for SL prediction often fail to integrate consistent and specific information across multiple biological networks.

Purpose of the Study:

• To develop a comprehensive representation learning framework for gene pairs that captures both multi-network consistency and network-specific information.
• To improve the accuracy of synthetic lethality prediction using computational approaches.

Main Methods:

• Proposed MGANSL (Multi-network consistent and specific representation with Generative Adversarial Network for Synthetic Lethality prediction).
• Employed network-aligned and network-specific encoding modules for comprehensive multi-network representations.
• Utilized cross-network and intra-network adversarial generation to capture consistent and specific gene pair information.

Main Results:

• MGANSL effectively captures both consistent cross-network and specific intra-network information for gene pairs.
• Demonstrated the superiority of MGANSL in synthetic lethality prediction through experiments on human datasets.
• Identified novel synthetic lethal associations with potential for anti-cancer drug development.

Conclusions:

• The proposed MGANSL method significantly outperforms existing approaches in synthetic lethality prediction.
• The identified synthetic lethal associations can guide the development of novel anti-cancer therapeutics.
• MGANSL provides a promising computational tool for identifying potential drug targets in cancer treatment.