Fig. 2.

The structures of the CNN model in PhaGCN for training (A) and encoding (B). R1, R2, R3 are the reference genomes used for training. C1, C2, C3 are the contigs that need to be encoded. In train mode, sequences will be fed to CNN to update parameters during back propagation. In encoding mode. The pre-trained CNN will be used to encode the sequences into numerical vectors. Then these vectors will be adopted as node features in the knowledge graph