search

narutoten520 / SCGDL

Spatial Clustering for Spatial Omics Data via Geometric Deep Learning
MIT License
6 stars 2 forks source link

Questions about DLPFC dataset #2

Open czwdcm opened 1 year ago

czwdcm commented 1 year ago

Hi,

It is a fantastic method for analyzing spatial transcriptomics!

I tried to cluster DLPFC samples with SCGDL following the tutorial you present. I found the outcomes are quite different from the results inthe paper. I think there may be a problem with the dataset I am using, can you share it with me? Thank you very much! The following picture is the result I got.

b3fe00ad1909f94fa190200210d37e4

The following picture is the dataset I used.

b9be58214855892f9bf1f9048a9eb0a
narutoten520 commented 1 year ago

Hi, thanks for your attention of SCGDL. I provide two methods to get the DLPFC dataset and the ground truth: 1) You can download the Raw data (spatial folder and h5_filtered gene matrix) from the spatialLIBD project. (https://github.com/LieberInstitute/spatialLIBD)

  1. The ground truth can access from the STAGATE package (https://drive.google.com/drive/folders/10lhz5VY7YfvHrtV40MwaqLmWz56U9eBP).
  2. The ground truth can also access from this link (https://github.com/LieberInstitute/HumanPilot/tree/master/outputs/SpatialDE_clustering). The value counts of each layer in these two ground truth is the same, however, they may have the different clustering accuracy. (my own experiment experience)

For SCGDL, because i did not fix seed in the BGMM model, so you can run several times for this model to see the best performance. Thanks

wenhuidu commented 1 year ago

Hi, I also have the same problem. My DLPFC dataset was downloaded from GraphST, and I did not reproduce the results provided in your paper when I used your method. What is the reason?

LXKBIGER commented 1 year ago

Excuse me,I have the same question,too. It's not understandable of that how the raw dataset to include the spatial information for the need of the model. The outcomes I performed can't reach the SOTA in this article,so may I have your data '151676_Anndata.h5ad' or the code tutorial to get this data for reproducing?Thanks a lot.

zhubbbzhu commented 7 months ago

Hi, I also have the same problem. My DLPFC dataset was downloaded from GraphST, and I did not reproduce the results provided in your paper when I used your method. What is the reason?