SLGNN: synthetic lethality prediction in human cancers based on factor-aware knowledge graph neural network

Area of Science:

• Genetics
• Computational Biology
• Cancer Research

Background:

• Synthetic lethality (SL) is a genetic interaction selectively lethal to cancer cells.
• Computational methods are increasingly used to predict SL interactions due to experimental limitations.
• Existing methods often lack a deep understanding and interpretability of SL mechanisms.

Purpose of the Study:

• To propose a novel computational method, SLGNN, for predicting synthetic lethality (SL) gene pairs.
• To enhance the interpretability of SL predictions by modeling underlying biological factors.
• To improve the accuracy and biological relevance of SL interaction predictions.

Main Methods:

• SLGNN models SL interactions by defining and incorporating SL-related factors derived from gene knowledge graphs.
• Gene embeddings are generated through explicit message aggregation on the knowledge graph.
• Factor-based message aggregation on a constructed SL graph refines gene embeddings for supervised prediction.

Main Results:

• The SLGNN model accurately predicts SL interaction pairs.
• SLGNN provides enhanced interpretability by modeling gene preferences for SL-related factors.
• Experimental results demonstrate that SLGNN outperforms current state-of-the-art SL prediction methods.

Conclusions:

• SLGNN offers a powerful and interpretable approach for predicting synthetic lethality.
• The method advances the discovery of targeted cancer therapies by improving SL pair identification.
• SLGNN's superior performance and interpretability benefit both computational biologists and clinical researchers.