SPACEL.Splane.model.init_model

SPACEL.Splane.model.init_model(expr_ad_list, n_clusters, k=2, use_weight=True, train_prop=0.5, n_neighbors=6, min_prop=0.01, lr=0.003, l1=0.01, l2=0.01, latent_dim=16, hidden_dims=64, gnn_dropout=0.8, simi_neighbor=1, use_gpu=None, seed=42)

Initialize Splane model.

Build the model then set the data and paratemters.

Parameters:

  • expr_ad_list (list) – A list of AnnData object of spatial transcriptomic data as the model input.

  • n_clusters (int) – The number of cluster of the model ouput.

  • k (int) – The order of neighbors of a spot for graph construction.

  • use_weight – If True, the cell type proportion of Moran was used as the weight of the loss.

  • train_prop (float) – The proportion of training set.

  • n_neighbors – The number of neighbors for graph construction.

  • lr (float) – Learning rate of training.

  • latent_dim (int) – The dimension of latent features. It equal to the number of nodes of bottleneck layer.

  • hidden_dims (int) – The number of nodes of hidden layers.

  • gnn_dropout (float) – The dropout rate of GNN model.

  • simi_neighbor – The order of neighbors used for similarity loss. If is None, It equal to the order used for constructed graph.

  • seed – Random number seed.

Return type:

SplaneModel

Returns:

A DataFrame contained deconvoluted results. Each row representing a spot, and each column representing a cell type.