clustering.yaml Reference

This configuration file corresponds to --option=clustering and controls low-dimensional embedding, neighbor graph construction, and clustering.

Parameter	Default	Description
option	clustering	Fixed identifier for the analysis stage
results_folder	results	Root directory for analysis outputs
data_fold	data	Root directory for raw input data
sample_list	sample.txt	Path to the sample list file
run_type	visium	Platform type
channel	compare_analysis	Analysis channel
tsene	False	Whether to generate an additional tSNE plot
MIN_DIST	0.3	UMAP min_dist parameter
SPREAD	2	UMAP spread parameter
cluster_algorithm	leiden	Clustering algorithm selection
resolution	0.5	Resolution for Leiden or Louvain clustering
n_clusters	15	Number of clusters for KMeans
n_comps	20	Number of principal components used for dimensionality reduction
k_geom	15	BANKSY geometric neighbor parameter
max_m	1	BANKSY neighborhood order
nbr_weight_decay	scaled_gaussian	Strategy for neighborhood weight decay
lambda_list	0.2	Weight for spatial enhancement
sketch	False	Whether to use the sketched object for clustering label propagation
pcs	25	Number of PCs used in clustering
NEIGHBORS	10	Number of neighbors used for the neighbor graph

Tuning suggestions

1. In a standard workflow, start with cluster_algorithm=leiden and then tune resolution and pcs.

2. If spatial continuity is important, explore spatial enhancement settings such as k_geom and lambda_list.

3. If sketch was enabled during preprocess, it should also be enabled at this stage.