Proseg crashes or generates poor segmentation with Xenium 5k, CosMx 6k, etc

[yihming]

Hello,

I was trying Proseg v1.1.3 on our Xenium 5k dataset, which has 301,710,014 transcripts, 573,137 cells estimated initially, and 10,029 genes, and with 16 threads (my machine has 32 vCPUs):

Using 16 threads
Read 301710014 transcripts
     573137 cells
     10029 genes
Estimated full area: 78272410
Full volume: 2279683800
Using grid size 146.29108. Chunks: 5772

My process has been running for 4 days, and is still running. I just wonder if this is normal, or I should try some older version of Proseg? (It seems I should try v1.1.2+ as the dataset doesn't have transcripts.csv.gz but only transcripts.parquet and transcripts.zarr.zip).

Any suggestion would be appreciated. Thank you!

Sincerely, Yiming