hdWGCNA Algorithm

Purpose

Use the hdWGCNA algorithm to identify spatial gene co-expression gene sets.

Usage

SDAS coexpress hdwgcna -i st.h5ad -o outdir --bin_size 100 \
--input_layer raw_counts \
--selected_genes top5000  \
--moran_path ./moran.csv \
--n_cpus 8 \
--seed 42 \
--knn_neighbors 50  \
--max_shared_cells 15 \
--soft_power 8

Input Parameter Description

Parameter	Required	Default	Description
-i / --input	Yes		Stereo-seq h5ad, must contain raw expression matrix
-o / --output	Yes		Output directory
--bin_size	Yes	50	Bin size for resolution (20, 50, 100, 200, cellbin), consistent with input h5ad
--layer	No		Specify the layer of the raw expression matrix in h5ad (e.g. layers['raw_counts'])
--selected_genes	No	top5000	Selected gene mode : full(all the genes), topn(top n genes in Moran'I index)
--moran_path	No		Path to the precomputed Moran'I index csv file
--n_cpus	No	8	Number of parallel jobs for a speedup on multi-core machines
--seed	No	42	Random seed
--knn_neighbors	No	50	construct metacells: Number of neighboring cells covered by KNN algorithm
--max_shared_cells	No	15	construct metacells:maximum number of shared cells between two metacells
--soft_power	No		Used in network construction, by default automatically selects the lowest soft_power with a scale-free topology model fit of 0.8

Output Results Display

Result File	Description
<input_name>_hdwgcna.module.csv	The result csv of spatial highly variable genes (gene symbol+gene id) corresponding to the co-expression gene set (module)
<input_name>_hdwgcna.module_score.csv	The result csv of gene set scoring for co-expression gene sets
<input_name>_hdwgcna.coexpress.rds	rds file containing the results of co-expression gene sets
<input_name>_hdwgcna.module_score.png/pdf	Spatial heatmap of gene set scoring for co-expression gene sets
<input_name>_hdwgcna.all_coex_dendrogram.png/pdf	Dendrogram of similarity between co-expression gene sets
<input_name>_hdwgcna.softpowers.png/pdf	Bar chart of soft_power values for network construction
<input_name>_hdwgcna.moran.csv	If topn is used for calculation, outputs all gene Moran index and P values

• Result csv of co-expression gene sets: <input_name>_hdwgcna.module.csv, separated by commas. hdWGCNA output shows the spatial highly variable genes identified and their corresponding co-expression gene sets (modules). kME indicates the correlation strength between a gene's expression pattern and the module eigengene (Module Eigengene, ME) of the module it belongs to. The closer the kME value is to 1 or -1, the more likely the gene is a hub gene.

real_gene_name	geneid	Module	color	kME_Module1	kME_Module2	kME_grey	kME_Module3	kME_Module4	kME_Module5	kME_Module6	kME_Module7	kME_Module8	kME_Module9
A2M	ENSG00000175899	Module1	green	0.47946868988301	-0.107096403482606	-0.178114022165641	0.0676792398874597	0.095966109797419	-0.0907050325056857	-0.0529390531160642	-0.150612945887371	0.0878907827651177	0.0249952108382643
A2M-AS1	ENSG00000237094	Module1	green	0.54370397007705	-0.150011910577089	-0.254597937099371	0.0926882061841318	0.140032173496191	-0.115227951266487	-0.101675353602963	-0.222107282189061	0.0803636102659976	0.0426306888623326
A2ML1	ENSG00000166535	Module2	yellow	0.0404144692736028	0.479908573141937	0.194701680726881	-0.327610748128114	0.0430624759042059	0.429681007497005	-0.342984504779987	0.145625804577339	-0.386999928188458	0.08281144751312791
A2MP1	ENSG00000256069	grey	grey	-0.046660656715667	0.20294339804614	0.284819067476003	-0.0506850476403686	-0.205976941174478	0.244779685854094	0.000250607520833238	0.170101997387916	-0.0177549796818324	0.0639042087827032

• Result csv of gene set scoring for co-expression gene sets: <input_name>_hdwgcna.module_score.csv, separated by commas. hdWGCNA output shows the high and low expression scores of each co-expression gene set (module).

	Module6	Module3	Module8	Module2	grey	Module7	Module5	Module9	Module1	Module4
2200_16100	-3.23688863476392	-4.34756288337066	-2.3278151796256	-8.21694142422341	-14.8112682710791	-9.12253218247156	-10.174563894144	-3.09447240000024	0.481660736850741	3.91787079378259
2200_17200	5.77873502485046	0.783016254503074	1.06582091429724	-6.03050203635639	-3.71256039305597	-0.825856084852031	-3.67468239887104	-2.09159016878048	-2.639251117267012	5.41583186417414
2300_16700	7.90521666109811	2.93759207152763	-0.391450035802177	-3.02639637030598	1.63013439679168	1.66371621513915	-1.51360146647437	-0.8975499248414	-4.66703690157902	1.40723191567521

• Spatial heatmap of gene set scoring for co-expression gene sets <input_name>_hdwgcna.module_score.png/pdf: Visualizes the spatial distribution patterns of all co-expression gene sets (Modules). The color intensity in the figure indicates the expression level of the co-expression gene set.

• Bar chart of soft_power values for network construction <input_name>_hdwgcna.softpowers.png/pdf: Analyzes the effect of different soft_power parameters on network construction. By default, the lowest soft_power with a scale-free topology model fit of 0.8 is automatically selected.

• Dendrogram of similarity between co-expression gene sets <input_name>_hdwgcna.all_coex_dendrogram.png/pdf: Shows the hierarchical clustering dendrogram of similarity between different co-expression gene sets (Modules).

Result Interpretation

• The co-expression gene sets start from Module1, and grey represents genes that do not meet the clustering requirements of co-expression gene sets.

Parameter Tuning Suggestions

• If the number of genes in bin20/50 samples is less than 200, or for other special samples, and the identified spatial co-expression gene sets are few, you can try lowering the threshold according to the soft_power test chart.

• You can customize the parameters knn_neighbors and max_shared_cells to obtain more interpretable results.