scellrun — full pipeline report

Run directory: run-20260430-151851

At a glance

QC pass rate: 87.0% (10835 / 12451 cells)
Integration method: none
Annotation panel: chondrocyte_markers
Chosen resolution: 0.313 clusters
Top cell-type labels:

Decision summary

27 decisions logged across 5 stage(s). Rows shaded amber were user overrides; ai tags an AI-driven choice.

qc
sourcekeyvaluedefaultrationale
user profile joint-disease default profile 'joint-disease' chosen; profiles ship tissue-tuned thresholds (mt%, hb%) and marker panels.
auto assay scrna fresh-tissue scRNA defaults; mt% ceiling 20% from PI cohort
auto species human drives mt/ribo/hb gene-pattern matching (uppercased; same regex covers MT-CO1 and mt-Co1)
auto max_pct_mt 20.0 mt% ceiling 20.0% — joint tissue is stress-prone, the textbook 10% silently drops real chondrocytes (PI cohort 2024-2026, AIO PM=20)
auto min_genes 200 min n_genes 200 — Ilicic 2016 / Luecken & Theis 2019 droplet-vs-cell floor
auto max_genes 4000 max n_genes 4000 — multiplet upper cap on 10x v3 chemistry; raise via --max-genes for high-RNA cell types (megakaryocytes, hepatocytes)
auto flag_doublets True scrublet on; report score distribution, never auto-drop (AIO leaves DF result attached)
auto raw_counts_check looks_like_raw input matrix sampled (200 rows): integer-valued + max>1 = looks_like_raw; non-integer with small max = looks_like_normalized; otherwise unknown
auto lang en report language; both EN and ZH templates kept in sync
integrate
sourcekeyvaluedefaultrationale
auto method none harmony --method 'none' chosen explicitly
auto sample_key none no sample/batch column in obs — single-batch run
auto n_pcs 30 PCA components for neighbours/UMAP; 30 covers most tissue heterogeneity
auto resolutions [0.1, 0.3, 0.5, 0.8, 1.0] Leiden multi-resolution sweep; 5 values (default short sweep)
auto regress_cell_cycle False off by default; turn on with --regress-cell-cycle if cycling cells dominate
markers
sourcekeyvaluedefaultrationale
auto resolutions [0.1, 0.3, 0.5, 0.8, 1.0] auto-detected from leiden_res_* columns in the integrated h5ad
auto logfc_threshold 1.0 |log2fc| >= 1.0 — Seurat in-house default; tightens to 1.0 to drop near-zero-effect markers
auto pct_min 0.25 min fraction expressing >= 0.25 — drops genes hit by dropout in the focal cluster
auto only_positive True positive-only markers (cluster-up) — what FindAllMarkers ships by default
annotate
sourcekeyvaluedefaultrationale
user profile joint_disease default profile selects which marker panels are available for matching
user panel chondrocyte_markers --panel 'chondrocyte_markers' forced
auto resolution 0.3 matching against leiden_res_0.3 clusters; the orchestrator picks this from integrate's quality table when called via `scellrun analyze`
auto use_ai False AI second-opinion off — deterministic panel match only
auto use_pubmed False PubMed lookup off — turn on with --pubmed for the literature evidence column
user tissue OA cartilage tissue context 'OA cartilage' drives PubMed scoping and AI prompt
analyze
sourcekeyvaluedefaultrationale
auto method_downgrade none harmony no sample/batch column (orig.ident/sample/batch/donor) in obs — single-sample input; auto-downgraded --method from harmony to none. Pass --method harmony explicitly to force the original behavior.
auto chosen_resolution_for_annotate 0.3 fewest singletons → most balanced (every resolution fragmented) — picked res=0.3: n_clusters=13, largest=31.5%, smallest=0.2%, singletons=2
auto annotate.auto_panel chondrocyte_markers fine-subtype panel preferred (chondrocyte hits dominate or are tied)

Stages

1. qc 01_qc/open report.html
2. integrate 02_integrate/open report.html
3. markers 03_markers/open report.html
4. annotate 04_annotate/open report.html

Run manifest (chronological)

#commandran atkey params
1 analyze:qc 2026-04-30T15:20:32.649120+00:00 profile=joint-disease species=human thresholds={'species': 'human', 'min_genes': 200, 'max_genes': 4000, 'min_counts': 500, 'max_pct_mt': 20.0, 'max_pct_ribo': 50.0, 'max_pct_hb': 2.0, 'min_cells_per_gene': 3, 'flag_doublets': True} force=False lang=en attempt_id=cae89793d0f9470a9c7f38894928f304
2 analyze:integrate 2026-04-30T15:20:52.573352+00:00 method=none sample_key=None n_pcs=30 resolutions=[0.1, 0.3, 0.5, 0.8, 1.0] regress_cell_cycle=False species=human use_ai=False tissue=OA cartilage lang=en attempt_id=cae89793d0f9470a9c7f38894928f304
3 analyze:markers 2026-04-30T15:23:51.773654+00:00 resolutions=[0.1, 0.3, 0.5, 0.8, 1.0] logfc_threshold=1.0 pct_min=0.25 only_positive=True top_n=10 lang=en attempt_id=cae89793d0f9470a9c7f38894928f304
4 analyze:annotate 2026-04-30T15:24:49.548069+00:00 profile=joint-disease panel_name=chondrocyte_markers resolution=0.3 use_ai=False use_pubmed=False tissue=OA cartilage lang=en attempt_id=cae89793d0f9470a9c7f38894928f304

Provenance trail

Each stage's report carries three tiers of evidence:

Generated by scellrun report. To export to PDF, open this page in a browser and use Print → Save as PDF (Chromium prints these layouts cleanly).