Synthetic lethality (SL) is a genetic interaction that refers to the phenomenon of cell death caused by the simultaneous inactivation of two non-lethal genes. Due to high-cost constraints and time consumption of experimental screening, computational prediction methods have become the main research tool. Currently, methods based on machine learning have been widely used in SL research, and discovering effective features to enhance the accuracy of predictions remains the key challenge to overcome in current research. We propose an SL prediction method based on graph embedding. First, we transformed five types of raw omics data into graph structures to capture the complex associations among genes. Then, using the graph embedding technique, we extracted feature information for each gene and constructed the feature representation of SL pairs by mathematical operations. Finally, different from GNN, which infers a single graph, we used the machine learning classifiers to discriminate positive and negative samples. Our method achieved better AUC than GNN-based baseline methods. Overall, this study firstly proposed a prediction model for Escherichia coli (E. coli) SLs that integrates the advantages of graph embedding techniques and classifier ensembles, which significantly improves the accuracy and reliability of prediction, and also provides new perspectives and methods for this field.
Feng et al. (Thu,) studied this question.