torchcell.models.dango.MetaEmbedding

class MetaEmbedding(hidden_channels: int)

Bases: Module

Meta-embedding module to integrate embeddings from multiple networks.

As described in the paper, this module: 1. Takes embeddings from 6 different PPI networks for each node 2. Uses an MLP to compute attention weights for each embedding 3. Combines embeddings using a weighted sum based on learned attention

forward(embeddings_dict: dict[str, Tensor]) → Tensor

Forward pass to integrate embeddings from multiple networks

Parameters:

embeddings_dict – Dictionary mapping edge types to node embeddings Each value has shape [num_nodes, hidden_channels]

Returns:

Integrated embeddings with shape [num_nodes, hidden_channels]