Open UnkownNames opened 7 months ago
And I found the result of the demo is different between two runs, may be there are some random seed need to be fixed, run one:
run two:
Hi @UnkownNames ,
Thank you for trying iStar. For the two issues you mentioned:
The sample in the demo data corresponds to Section G1 in Andersson et al., 2021.
Could you check whether your clustering results (data/humam/02/clusters-gene/labels.png
) are identical in your two runs? It is often the case that clustering results from multiple runs on the same data are very similar (even if some clusters might be swapped) but not exactly the same. I have noticed that the k-means algorithm used for clustering could give different results even if the same seed is used. I am working on a solution to this problem.
Dear Daiwei,
Firstly, congrats to you for developing this perfect tool!
I also have the same question as required by UnkownNames .
You mentioned that:
- The sample in the demo data corresponds to Section G1 in Andersson et al., 2021.
But, when I checked this dataset about Section G1, I find the gene expression file named "G1.tsv.gz" "her2st/data/ST-cnts /G1.tsv.gz", this is the file named "cnts.tsv" in iSTAR.
Then, I can't find another input file named "locs-raw.tsv" and its corresponding "H&E image" file. So, could you give me more details about your input files in this demo. I only found the file named "G1_selection.tsv" her2st/data/ST-spotfiles /G1_selection.tsv, followed by the gene expression file "G1.tsv.gz". I noticed that it is different from yours. Just like these two figures:
In addition, could you provide some dataset or information about your results about Fig. 1b. I care about how you get the results using Xenium data, expect an example when you are available. I would greatly appreciate it if you could send me (sdnully2012@163.com, lingyuli@hku.hk) the processed data or code.
Thanks a lot.
Lingyu Li
Great work, I have run the
run_demo.sh
successfully. But when I want to reproduce some result displayed in the paper, I found the raw histology image used in the demo is not matched in any histology image in the paper, would you provide more information about this dataset and the running result in the demo? In addition, the iStar would be more friendly to users if you organize code into packages(can be submitted to pypi) and provide a more detailed tutorial(like a file of ipynb with jupyter notebook)? Thank you!