KR4SL: knowledge graph reasoning for explainable prediction of synthetic lethality.

Area of Science:

• Computational biology
• Genomics
• Bioinformatics

Background:

• Synthetic lethality (SL) is a key anticancer strategy, but wet-lab screening is costly and prone to off-target effects.
• Knowledge graphs (KGs) can improve machine learning-based SL prediction, yet subgraph structures remain underexplored.
• Lack of interpretability in current machine learning methods hinders their application in SL identification.

Purpose of the Study:

• To develop a novel computational model, KR4SL, for predicting synthetic lethality (SL) partners.
• To leverage knowledge graph (KG) structural and textual semantics for enhanced SL prediction.
• To improve the interpretability of SL prediction models by identifying critical subgraph structures.

Main Methods:

• KR4SL constructs and learns from relational digraphs within a KG, fusing entity textual semantics.
• A recurrent neural network enhances sequential path semantics, while an attentive aggregator identifies explanatory subgraphs.
• The model predicts SL partners by capturing structural and semantic information from KGs.

Main Results:

• KR4SL significantly outperforms existing baseline methods in predicting synthetic lethality partners.
• The model provides interpretable explanations through identified subgraph structures, revealing underlying mechanisms.
• Experimental results demonstrate KR4SL's superior predictive power and interpretability.

Conclusions:

• KR4SL offers a powerful and interpretable deep learning approach for identifying synthetic lethality targets.
• The model's ability to uncover mechanisms enhances its utility in cancer drug target discovery.
• This work highlights the practical value of deep learning and KG exploration in advancing SL-based cancer therapies.