KG-SLomics: Synthetic Lethality Prediction Using Knowledge Graph and Cancer Type-Specific Multiomics Integrated Graph...

Area of Science:

• Genomics
• Computational Biology
• Cancer Research

Background:

• Synthetic lethality (SL) exploits the simultaneous alteration of two genes to induce cancer cell death.
• PARP inhibitors showcase SL's potential for targeting cancers with undruggable mutations.
• Existing SL prediction methods face challenges with cancer type variability and limited data integration.

Purpose of the Study:

• To develop a novel computational model for predicting synthetic lethality pairs.
• To address the limitations of existing methods in generalizing across diverse cancer types.
• To identify novel therapeutic targets for undruggable cancer mutations.

Main Methods:

• Developed KG-SLomics, a relational graph attention network model.
• Constructed a comprehensive knowledge graph (KG) with updated biological entities.
• Integrated pre-trained KG embeddings with multiomics data for topological and cancer-specific feature capture.

Main Results:

• KG-SLomics achieved high accuracy in predicting SL probabilities.
• The model demonstrated superior performance compared to advanced baseline methods.
• High attention scores were allocated to relevant biological entities within the KG.

Conclusions:

• KG-SLomics offers a robust and generalizable approach for synthetic lethality prediction.
• The model successfully identified potential novel therapeutic targets for cancer treatment.
• This work highlights the clinical potential of knowledge graph-enhanced multiomics data analysis in precision oncology.