Step 1: find_latent_representation¶
usage: gsmap run_find_latent_representations [-h] --workdir WORKDIR
--sample_name SAMPLE_NAME
--input_hdf5_path INPUT_HDF5_PATH
--data_layer DATA_LAYER
[--annotation ANNOTATION]
[--epochs EPOCHS]
[--feat_hidden1 FEAT_HIDDEN1]
[--feat_hidden2 FEAT_HIDDEN2]
[--gat_hidden1 GAT_HIDDEN1]
[--gat_hidden2 GAT_HIDDEN2]
[--p_drop P_DROP]
[--gat_lr GAT_LR]
[--n_neighbors N_NEIGHBORS]
[--n_comps N_COMPS]
[--weighted_adj]
[--convergence_threshold CONVERGENCE_THRESHOLD]
[--hierarchically]
[--pearson_residuals]
Named Arguments¶
- --workdir
Path to the working directory.
- --sample_name
Name of the sample.
- --input_hdf5_path
Path to the input HDF5 file.
- --data_layer
Data layer for gene expression (e.g., “count”, “counts”, “log1p”).
Default:
'counts'
- --annotation
Name of the annotation in adata.obs to use.
- --epochs
Number of training epochs.
Default:
300
- --feat_hidden1
Neurons in the first hidden layer.
Default:
256
- --feat_hidden2
Neurons in the second hidden layer.
Default:
128
- --gat_hidden1
Units in the first GAT hidden layer.
Default:
64
- --gat_hidden2
Units in the second GAT hidden layer.
Default:
30
- --p_drop
Dropout rate.
Default:
0.1
- --gat_lr
Learning rate for the GAT.
Default:
0.001
- --n_neighbors
Number of neighbors for GAT.
Default:
11
- --n_comps
Number of principal components for PCA.
Default:
300
- --weighted_adj
Use weighted adjacency in GAT.
Default:
False
- --convergence_threshold
Threshold for convergence.
Default:
0.0001
- --hierarchically
Enable hierarchical latent representation finding.
Default:
False
- --pearson_residuals
Using the pearson residuals.
Default:
False