How to get the centrality encoding module and spatial encoding module from gene graph

[LeeWang21]

Hi, developer!
Thanks for developing the tools. The CMM pre-training task is impressive, especially the GeneGraph attention layers. But I am confused about the calculation of centrality encoding module and spatial encoding module.

For centrality encoding module:
In the source code, I see the in_degree_embed and out_degree_embed , which are just the nn.Embeddings.

self.embed_in_degree = nn.Sequential(
            nn.Embedding(20000, self.d_model, padding_idx=0)
        )
self.embed_out_degree = nn.Sequential(
            nn.Embedding(20000, self.d_model, padding_idx=0)
        )

is this the centrality encoding module? It seems that the embed_in_degree and embed_out_degree are not related to gene graph.

For spatial encoding module, I do not known the source code.

could you plz tell me how to get the centrality encoding module and spatial encoding module from gene graph ?

Best wishes

[TianzeCompbio]

Hi, LeeWang21,

Thanks for your interest in the CellNavi source code! The nn.Embedding layers you saw (embed_in_degree and embed_out_degree) are indeed part of the centrality encoding module — but they don’t directly “compute” anything from the gene graph. They’re lookup tables that transform the pre-computed in/out degree of each node (from the gene graph) into learnable embedding vectors. For more details, please refer to the official PyTorch documentation for nn.Embeddings.