Concepts & Terminology¶

Parameters for GNN model used in ONTraC¶

The training process will terminated if the model does not improve after the number of epochs set by this parameter.

The model will be considered improved if the total loss decreases by the propotion set by this parameter.

The number of niche features after GCN processing (step2).

The modularity loss controls the spatial smoothness of niche clusters.

The purity loss controls the purity of cell type composition within each niche cluster.

The regularization loss controls the balance of niche cluster sizes. The higher this weight is set, the more equal the size of each niche cluster.

The beta value of the softmax function used in generating niche cluster assignment matrix.