How do I actually use an external model?

[jeremymchacon]

Hello,

I am trying to use an external model so that labels are consistent across samples. I have ficture running successfully without the external model, making figures, etc. However, the below code does not produce any figures. Please trust that the inputs are accurate because they work in a normal (-all) ficture run. Below the code I show the top of my input_model and input_cmap.

ficture run_together --in-tsv ${input_dir}/${sample_file} \
 --in-minmax ${input_dir}/${minmax_file} \
 --in-feature ${input_dir}/unfitered_gene_list.txt.gz \
 --mu-scale 1 \
 --major-axis X \
 --key Count \
 --decode-from-external-model \
 --external-model ${input_model} \
 --external-cmap ${input_cmap} \
 --n-jobs 8 \
 --plot-each-factor \
 --out-dir ${output_dir}

Here is the command output. Note that it successfully makes batch.matrix.tsv, but does not put any files into the analysis subdirectory (just a folder with the name of my $input_model, with an empty folder called figure inside that).

--------------------------------------------------------------
Creating minibatch from /home/perlinge/shared/RIS_analysis/cosMX_MD/input//unfiltered_tx_data_for_6W_L_tibialis.txt.gz...
--------------------------------------------------------------
ficture make_spatial_minibatch --input /home/perlinge/shared/RIS_analysis/cosMX_MD/input//unfiltered_tx_data_for_6W_L_tibialis.txt.gz --output /home/perlinge/shared/RIS_analysis/cosMX_MD/output//mouseref_6W_L_tibialis/batched.matrix.tsv --mu_scale 1.0 --batch_size 500 --batch_buff 30 --major_axis X
INFO:root:Random seed 1747406156.032602
['X', 'random_index', 'Y', 'gene', 'Count']
INFO:root:Read blocks of pixels: 2038.90 x 3390.28
INFO:root:Output region (501.00, 501.00) (4998.1, 5499.1) x (0.0, 501.0)
INFO:root:Output region (501.00, 500.00) (4998.1, 5499.1) x (471.0, 971.0)
INFO:root:Output region (500.00, 501.00) (5469.1, 5969.1) x (0.0, 501.0)
INFO:root:Output region (500.00, 500.00) (5469.1, 5969.1) x (471.0, 971.0)
INFO:root:Output region (500.00, 500.00) (5469.1, 5969.1) x (941.0, 1441.0)
INFO:root:Output region (500.00, 500.00) (5469.1, 5969.1) x (1881.0, 2381.0)
INFO:root:Output region (500.00, 500.00) (5469.1, 5969.1) x (2351.0, 2851.0)
INFO:root:Output region (500.00, 501.00) (5939.1, 6439.1) x (0.0, 501.0)
INFO:root:Output region (500.00, 500.00) (5939.1, 6439.1) x (471.0, 971.0)
INFO:root:Output region (500.00, 500.00) (5939.1, 6439.1) x (941.0, 1441.0)
INFO:root:Output region (500.00, 500.00) (5939.1, 6439.1) x (1411.0, 1911.0)
INFO:root:Output region (500.00, 500.00) (5939.1, 6439.1) x (1881.0, 2381.0)
INFO:root:Output region (500.00, 500.00) (5939.1, 6439.1) x (2351.0, 2851.0)
INFO:root:Output region (628.90, 501.00) (6409.1, 7038.0) x (0.0, 501.0)
INFO:root:Output region (628.90, 500.00) (6409.1, 7038.0) x (471.0, 971.0)
INFO:root:Output region (628.90, 500.00) (6409.1, 7038.0) x (941.0, 1441.0)
INFO:root:Output region (628.90, 500.00) (6409.1, 7038.0) x (1411.0, 1911.0)
INFO:root:Output region (628.90, 500.00) (6409.1, 7038.0) x (1881.0, 2381.0)
INFO:root:Output region (628.90, 500.00) (6409.1, 7038.0) x (2351.0, 2851.0)
INFO:root:Output region (628.90, 570.28) (6409.1, 7038.0) x (2821.0, 3391.3)
INFO:root:Left over size 34212 (34212, 30.00)
INFO:root:Read blocks of pixels: 1340.91 x 3216.92
INFO:root:Output region (501.00, 501.00) (7007.0, 7508.0) x (173.4, 674.4)
INFO:root:Output region (501.00, 500.00) (7007.0, 7508.0) x (644.4, 1144.4)
INFO:root:Output region (501.00, 500.00) (7007.0, 7508.0) x (1114.4, 1614.4)
INFO:root:Output region (501.00, 500.00) (7007.0, 7508.0) x (1584.4, 2084.4)
INFO:root:Output region (501.00, 500.00) (7007.0, 7508.0) x (2054.4, 2554.4)
INFO:root:Output region (501.00, 500.00) (7007.0, 7508.0) x (2524.4, 3024.4)
INFO:root:Output region (501.00, 396.92) (7007.0, 7508.0) x (2994.4, 3391.3)
INFO:root:Output region (500.00, 501.00) (7478.0, 7978.0) x (173.4, 674.4)
INFO:root:Output region (500.00, 500.00) (7478.0, 7978.0) x (644.4, 1144.4)
INFO:root:Output region (500.00, 500.00) (7478.0, 7978.0) x (1114.4, 1614.4)
INFO:root:Output region (500.00, 500.00) (7478.0, 7978.0) x (1584.4, 2084.4)
INFO:root:Output region (500.00, 500.00) (7478.0, 7978.0) x (2054.4, 2554.4)
INFO:root:Output region (500.00, 500.00) (7478.0, 7978.0) x (2524.4, 3024.4)
INFO:root:Output region (500.00, 396.92) (7478.0, 7978.0) x (2994.4, 3391.3)
INFO:root:Output region (400.91, 501.00) (7948.0, 8348.9) x (173.4, 674.4)
INFO:root:Output region (400.91, 500.00) (7948.0, 8348.9) x (644.4, 1144.4)
INFO:root:Output region (400.91, 500.00) (7948.0, 8348.9) x (1114.4, 1614.4)
INFO:root:Output region (400.91, 500.00) (7948.0, 8348.9) x (1584.4, 2084.4)
INFO:root:Output region (400.91, 500.00) (7948.0, 8348.9) x (2054.4, 2554.4)
INFO:root:Output region (400.91, 500.00) (7948.0, 8348.9) x (2524.4, 3024.4)
INFO:root:Output region (400.91, 396.92) (7948.0, 8348.9) x (2994.4, 3391.3)
INFO:root:Left over size 8662 (8662, 30.00)
INFO:root:Read blocks of pixels: 302.05 x 1027.42
INFO:root:Read blocks of pixels: 302.05 x 1027.42
INFO:root:Output region (303.05, 501.00) (8317.9, 8620.9) x (1343.1, 1844.1)
INFO:root:Output region (303.05, 557.42) (8317.9, 8620.9) x (1814.1, 2371.5)
gzip -f /home/perlinge/shared/RIS_analysis/cosMX_MD/output//mouseref_6W_L_tibialis/batched.matrix.tsv

Here is the top of my reference. To make it, I took the reference from celldex::mouseRNAseqData, averaged to one column per cell type, and subsetted only the rows that had matching genes in my assay. I made sure it is tab-separated:

gunzip celldex_mouseRNAseqData.txt.gz
cut 
-f1,2,3,4 celldex_mouseRNAseqData.txt | head
gene Adipocytes Neurons Astrocytes
Aatk 5.767391 7.477957 8.565632
Abl1 8.209882 7.558083 8.362141
Abl2 7.370633 7.986000 8.856147
Acacb 10.247353 4.409268 7.413806
Ace 8.514046 5.445285 4.906251
Acp5 8.273683 3.711571 3.431362
Acta2 9.854108 5.133704 4.689625
Actg2 5.949027 4.141616 3.534786
Acvr1 6.827253 7.095488 7.197461

What I would like is the same outputs I get when I run it without the reference, and with the --all flag.

Thank you!!

[Yichen-Si]

Hi, I think you just need to pass --fit-width, perhaps with the size you used for --train-width before. If you were using the default value then pass --fit-width 12 (12 micron)
Apologies for the confusion.

[jeremymchacon]

Thanks, in the end I grabbed the hexagon... file, loaded it into seurat, and used singleR. Is there anything wrong with that approach from your perspective?

[Yichen-Si]

I've not used singleR before, I guess it is OK if you don't need pixel level information.

[jeremymchacon]

Am I interpreting the hexagon file correctly in that it contains the data as assigned to anchors? And then each random_index is a different anchor? What would be the additional data I would gain at the pixel level? Thanks!

…

On Fri, May 16, 2025 at 3:39 PM Yichen-Si ***@***.***> wrote: *Yichen-Si* left a comment (seqscope/ficture#45) <#45 (comment)> I've not used singleR before, I guess it is OK if you don't need pixel level information. — Reply to this email directly, view it on GitHub <#45 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ABLGCRPBW3RA4I3R77TSF6D26ZEG3AVCNFSM6AAAAAB5I6EMQSVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDQOBXGYZTMNJUHE> . You are receiving this because you authored the thread.Message ID: ***@***.***>

-- *Jeremy M. Chacón, Ph.D.* *he / him / his* *Bioinformatics Analyst* *RI Bioinformatics* *Minnesota Supercomputing Institute* *University of Minnesota*

[Yichen-Si]

No, the hexagon file just contains the data grouped into hexagons on a grid, no inference/analysis was done. You can try run_together with --decode-from-external-model and --fit-width 12 again.