Skip to content

Choose smoke, fast, or full

For Users choosing processing depth before inference
Hands-on steps Compare presets, select a separate output, dry-run, execute
Prerequisites Reviewed inspection manifest, verified source MPP, ready host/model
Download/storage Container/model downloads may occur; work/results increase with processed tissue
Writes to A distinct output-associated work/result pair per preset
Preset Mapping Purpose
smoke one selected slide, seeded 1%, fail fast First real technical check
fast seeded 10% by default Exploratory composition and iteration
full 100% of detected tissue tiles Exhaustive run after QC/resource review

Run a dry plan

tumorquantai run /data/slides \
  --output /data/results-smoke \
  --preset smoke \
  --source-mpp "$SOURCE_MPP" \
  --cpu \
  --dry-run

The CLI prints the expanded run.sh/Nextflow mapping with secrets redacted and states CPU/GPU selection, sampling, seed, result path, and work path.

Use exactly one of --cpu or --gpu when the execution path must be explicit. Omit both to retain automatic selection. Existing automation may continue to use --profile cpu, --profile gpu, --profile auto, or --profile local.

Execute by removing --dry-run. Use --sample SAMPLE_ID when inspection found multiple slides and you need a specific smoke slide.

Keep modes separate

/data/results-smoke/  (default work: /data/results-smoke/.tumorquantai-work/)
/data/results-fast/   (default work: /data/results-fast/.tumorquantai-work/)
/data/results-full/   (default work: /data/results-full/.tumorquantai-work/)

Do not reuse one result root across fast and full. Counts from different processed areas are not directly comparable. The main CLI refuses an unsafe accidental mixture.

Sampling is deterministic for the same input fingerprint, percentage, and seed. Fast counts are detected cells in sampled tissue tiles—not full-slide counts. Never multiply by 100 / percent_slide.

Stop, resume, and clean up

Press Ctrl+C, keep the work directory, and repeat the identical command. Use status for exact recovery instructions. Clean only the named result/work pair after verifying outputs and audit tables.

Next: review QC overlays after the smoke run.