integration hexagon count matrix and decoded factor matrix

[dahun73]

Hello,

Thank you for serving such a nice tool in advance :)

Actually, I want to make one integrated matrix for including decoded factor and expression count matrix(I want to visualize decoded factor and gene expression side-by-side following tissue coordinates). So, I tried to check all output files after running 'run_together'. In 'hexagon.d_12.tsv.gz' file, there are randon_index labels and gene expression count. However, I couldn't match these random_index labels with other output files. I also tried to find additional information in 'FICTURE' manual and here but there is no information of this issue. Are there any tips for integrating these data?

Thanks!
Dahun

[Yichen-Si]

Hi Dahun,

Sorry for the inconvenience. I guess you want a file where each row is a transcript/pixel, with both gene ID and decoded factors? Unfortunately in the current version we don't have an option to do that directly, but there are a few ways you can try:
1: use spatula: https://seqscope.github.io/spatula/tools/join_pixel_tsv/
2: do a nearest neighbor assignment so that each original pixel/transcript gets the factor assignment of its nearest neighbor in the decoded output. This might take some time and memory if you try to do it for all pixels.
3: try our new implementation in https://github.com/Yichen-Si/punkst . Documentation is here (in particular, in pixel-decode there is an optional flag --output-original to include gene and gene count information in the pixel level output.
It is probably too much trouble to install the new version and do the analysis again, so I hope either 1 or 2 works for you, but it might be the easiest way if your data is big and you might run FICTURE again on this or other data.

[dahun73]

Dear @Yichen-Si,

Thank you for your kind reply! However, I have some questions before running 'spatula join_pixcel_tsv'.

1. In 'Sorting the pixel-level output from FICTURE'(before running joun_pixel_tsv), if I ran FICTURE on 10x Visium HD data, would the major axis be the Y-axis?

2. For me, to run this 'join_pixel_tsv', I would put 'hexagon.d_12.tsv.gz' in the input tsv and 'nF24.d_12.decode.prj_12.r_4_5.pixel.sorted_by_major_axis.tsv.gz' in the pix-prefix-tsv? The manual doesn't have any guidelines for different types of spatial transcriptomics, which makes this command a little difficult to execute.

3. Finally, I was wondering if these files are correct for coordinates and factor information and expression analysed at the hexagonal level.
   expression count matrix : hexagon.d_12.tsv.gz
   coordinates and factor matrix : nF24.d_12.decode.prj_12.r_4_5.pixel.sorted.tsv.gz

In my case, nuclei segmentation was not working well in 10X visium-hd data. So for me, the FICTURE is fantastic. Yesterday, I was looking to see how I could do some further analysis after running FICTURE and I noticed that there was a new implementation. I should also use punkst tool as well. Thank you so much again.

Dahun

[Yichen-Si]

Hi Dahun,

Now I think I might have misunderstood what you were asking, could you tell me what you want the output to look like?
Since you've tried spatula, I guess you don't mind trying another c++ tool : ) then using our new implementation in punkst may be easier.

For your question about using spatula:
1 If you used run_together, your pixel level output is sorted by the Y-axis
2 The hexagon files collapse pixels to low resolution, they should not be used. spatula join-pixel-tsv is meant to combine the original pixel level output with FICTURE output, so the input should be the transcript file (the input to FICTURE).
3 No, the expression count matrix should be the transcript file (in this case, the 2um resolution data from Visium HD), that is the input to FICTURE.

[dahun73]

Dear @Yichen-Si,

Ahhhh!! I understood! So what you're saying is that the hexagon file is low resolution, so I should merge it with the original transcript file and the FICTURE pixel output.

What I actually want to do is I want to visualize other gene expression levels for each factor. Because some factors that were followed by FICTURE are interested, I want to check the other genes in these factors. So, I tried to integrate transcripts data and factor assigned pixel matrix in hexagon format. Actually, you said hexagon file collapsed so it's low resolution, right? But I think it is also okay if their sizes are not more than nuclei size(I also want matrix as the same size of nuclei).

Anyway, I should try 2um resolution transcripts file and FICTURE output first.

This might be wishful thinking, but I think it would be really helpful to others if the manual also showed how to apply it with other packages (like Seurat or Scanpy) to allow for further analysis.

Thank you so much again!

Dahun