Denoising gene expression by stCluster

In addition to clustering, the embeddings generated by stCluster can be applied to a variety of other downstream tasks. In this section, we will use gene expression denoising as an example to assess the expressive of the embeddings learned by stCluster.

import scanpy as sc
from st_datasets.dataset import get_data, get_dlpfc_data
from stCluster.eval import eval

adata, n_cluster = get_data(dataset_func=get_dlpfc_data, id='151507')
adata, g = eval(adata, radius=150, model_paras_path='/home/hanshu/stCluster_paras/DLPFC/DLPFC_0.pkl')

>>> INFO: Download dataset: 100%|██████████| 98.8M/98.8M [00:05<00:00, 17.7MB/s]

>>> INFO: dataset name: dorsolateral prefrontal cortex (DLPFC), slice: 151507, size: (4226, 33538), cluster: 7.(6.991s)
>>> INFO: Input size torch.Size([4226, 3000]).
>>> INFO: Graph contains 28996 edges, average 6.861 edges per node.
>>> INFO: Build graph success!
>>> INFO: Finish load model, begin to generate embedding and rebuild gene expression, input data size: (4226, 3000).
>>> INFO: Finish embedding generation process, please use the embedding to do downstream evaluation, total time: 1.544s

To facilitate your assessment of the gene expression denoising effectiveness, stCluster incorporates a built-in denoising model that is used to train a denoised gene expression matrix through the embeddings.

import scanpy as sc
from stCluster.denoising import train

adata = train(data=adata, spatial_graph=g)  # train denoised gene expression matrix

# draw plots
sc.pl.spatial(adata, color=['GFAP', 'HPCAL1', 'ENC1', 'MAP1B', 'TMSB10', 'MBP'], spot_size=150)
sc.pl.spatial(adata, color=['GFAP', 'HPCAL1', 'ENC1', 'MAP1B', 'TMSB10', 'MBP'], layer='denoised_gene', spot_size=150)

>>> INFO: Input size torch.Size([4226, 30]).

>>> INFO: Training: 100%|██████████| 600/600 [00:03<00:00, 187.93it/s]

>>> INFO: Finish gene denoising process, total time: 3.267s.