search

QuKunLab / SpatialBenchmarking

BSD 2-Clause "Simplified" License
106 stars 26 forks source link

Cannot reproduce the results in Figure 4 #14

Open TianyouLuo opened 1 year ago

TianyouLuo commented 1 year ago

Hi,

Thank you for the nice work! I am trying to reproduce the STARmap simulation results of Figure 4, but the RCTD results I obtained are slightly different from what you shared in the folder FigureData/Figure4/Dataset10_STARmap/Result_STARmap/RCTD_result.txt. Specifically, I am using following datasets as the input data: Spatial count: FigureData/Figure4/Dataset10_STARmap/Simulated_STARmap/combined_spatial_count.txt scRNA-seq reference: from the DataUpload.zip file in the shared google drive: DataUpload/Dataset10/scRNA_count.txt scRNA-seq reference annotation: FigureData/Figure4/Dataset10_STARmap/Rawdata/starmap_sc_rna_celltype.tsv

Am I using the correct input files as you used for the STARmap experiments in Figure 4? Or did you do further preprocessing before using them as inputs for deconvolution? I tried to follow BLAST_CelltypeDeconvolution.ipynb as well, but in there it is mentioned that for SpatialDWLS, RCTD, Seurat and SPOTlight, the input is a .h5seurat file, which I cannot find.

For further details, I used the above mentioned files as input for RCTD with default parameters, and comparing the results I got with the shared RCTD_results.txt, the mean difference across cell types are listed below: Astro Endo Excitatory.L2.3 Excitatory.L4 0.008530383 0.013151105 0.033019338 0.040755041 Excitatory.L5 Excitatory.L6 Inhibitory.Other Inhibitory.Pvalb 0.037881955 0.058106645 0.048388479 0.050582087 Inhibitory.Sst Inhibitory.Vip Micro Neuron.Other 0.072285689 0.017592593 0.006866524 0.012171693 Olig Other Smc 0.019501107 0.009123502 0.004113764

I am trying to understand what caused the difference. Thanks!

QuKunLab commented 1 year ago

Hi~ Thank you for your interest in our data associated with the paper. We provided the h5ad file for users to reproducing our results. So if you want to get the result for RCTD, you can used the R command: Convert("FigureData/Figure4/Dataset10_STARmap/Simulated_STARmap/starmap_sc_rna.h5ad", dest = "h5seurat", overwrite = TRUE) Convert("FigureData/Figure4/Dataset10_STARmap/Simulated_STARmap/starmap_spatial.h5ad", dest = "h5seurat", overwrite = TRUE) to obtain the h5seurat files as input.

And we have checked the result for RCTD, we have get the same result with what we shared in the folder. We don't know if different platforms may cause slightly different results. You can use the converted h5seurat file sa input to compare the result for RCTD.