Proseg v3.0.5 on very large tissues

[victormanna]

Happy to report that Proseg v3.0.5 has run successfully in all our samples🎉 including the Proseg → Baysor → Xenium Ranger workflow.

I also re-segmented some publicly available cancer and healthy colon datasets from 10x Genomics to keep things consistent. These tissues are 3 - 4× times bigger, the runs finishes successfully with Proseg 3.0.5..

1. Ovarian 407K cells Prime Panel, Proseg runtime - 308m 58s https://www.10xgenomics.com/datasets/ffpe-human-ovarian-cancer-data-with-human-immuno-oncology-profiling-panel-and-custom-add-on-1-standard. wget https://s3-us-west-2.amazonaws.com/10x.files/samples/xenium/3.0.0/Xenium_Prime_Ovarian_Cancer_FFPE_XRrun/Xenium_Prime_Ovarian_Cancer_FFPE_XRrun_outs.zip

2. Colorectal 585K cells v1 Panel (Cancer, pre-designed + add-on panel), Proseg runtime - 177m 22s https://www.10xgenomics.com/datasets/human-colon-preview-data-xenium-human-colon-gene-expression-panel-1-standard wget https://s3-us-west-2.amazonaws.com/10x.files/samples/xenium/1.6.0/Xenium_V1_hColon_Cancer_Add_on_FFPE/Xenium_V1_hColon_Cancer_Add_on_FFPE_outs.zip

Now for the Proseg → Baysor → Xenium Ranger workflow, it shows out of memory with 48 Cores, 196GB RAM thus I increased memory to 500GB, with 64 cores and still, the conversion fails ⚠️

With Proseg 2.0.5 this was no problem at all and the conversion was successful (needed 32 cores, 128GB RAM).

proseg-to-baysor proseg-output.zarr
--output-transcript-metadata proseg-to-baysor-transcript-metadata.csv
--output-cell-polygons proseg-to-baysor-cell-polygons.geojson

xeniumranger import-segmentation
--id ${patient}_baysor_proseg
--xenium-bundle /data/cephfs-1/home/users/mannas_c/work/Env_Backups/cac_analysis/Xenium_v1_RAW/$patient
--transcript-assignment proseg-to-baysor-transcript-metadata.csv
--viz-polygons proseg-to-baysor-cell-polygons.geojson
--units=microns

/var/spool/slurmd/job18271678/slurm_script: line 40: 2797177 Illegal instruction (core dumped) xeniumranger import-segmentation --id ${patient}_baysor_proseg --xenium-bundle /data/cephfs-1/home/users/mannas_c/work/Env_Backups/cac_analysis/Xenium_v1_RAW/$patient --transcript-assignment proseg-to-baysor-transcript-metadata.csv --viz-polygons proseg-to-baysor-cell-polygons.geojson --units=microns

Now another interesting case 3) Healthy colon public v1 dataset (Non-diseased, pre-designed + add-on panel) (275K cells, relatively large tissue) https://www.10xgenomics.com/datasets/human-colon-preview-data-xenium-human-colon-gene-expression-panel-1-standard

wget https://cf.10xgenomics.com/samples/xenium/1.6.0/Xenium_V1_hColon_Non_diseased_Add_on_FFPE/Xenium_V1_hColon_Non_diseased_Add_on_FFPE_outs.zip

Proseg 3.0.5 works ✅ Proseg → Baysor → Xenium Ranger workflow also works ✅ But the visualization in xenium explorer does not work ⚠️ (segmented cells take forever to load, I have left it for an hour and it was still loading). This was not the case with the original 10x segmentation.

What do you suspect is going on?