search

messcode / STAMarker

STAMarker: Identify Spatial Domain-specific Variable Genes via Ensemble Graph Attention Autoencoders
4 stars 4 forks source link

Did you apply the StaMarker for the other slices of DLPFC dataset? #9

Closed keyalone closed 3 months ago

keyalone commented 4 months ago

Hi: STAMarker is highly innovative, and its results are very convincing. The ensemble-based learning method offers a novel approach to spatial domain detection. I noticed that the authors applied STAMarker only to slice 151507 of the DLPFC datasets and it's grateful that the authors provide the tutorial for the reproduce experimental results of sclice 151507. I am currently interested in applying STAMarker to all slices of the DLPFC. I would greatly appreciate it if the authors could provide the code for the DLPFC dataset. Alternatively, could the authors offer guidances on the parameter settings for applying STAMarker to the all slices of DLPFC datasets? My email address is lihs@zjut.edu.cn. I look forward to the authors' response. Thank you.

Best regards.

messcode commented 3 months ago

Hi keyalone

I applied the STAMarker to all slices of the DLPFC data. I used the same settings across the entire dataset, except for n_classes, which was set to the number of layers corresponding to each slice. 

dataset:
  name: DLPFC
  type: 10xVisum
  data_dir: ../datasets/DLPFC/
  section: 151673
  n_classes: 7

preprocessing:
  rad_cutoff: 150
  n_top_genes: 3000
  show_net_stats: false

Alternatively, you can set the model.clustering, model.consensus_clustering, and train_classifiers to the correct number of classes. Additionally, I have re-implemented the consensus_clustering using linear sum assignment to accelerate it. I also added functions to analyze the data (see T2 Human DLPFC 10x Visum dataset). I hope this helps.

keyalone commented 3 months ago

Hi messcode: Thank you for your prompt reply. Your answer is helpful.

Best regards.