Know your cell metadata before you analyze.
Every AnnData carries an obs table1 — one row per cell or sample, one column per label. The Obs Inspector node opens it up so you can browse each column, read its value counts and statistics, try a query against it, and make small, recorded edits — all without leaving Oreumics.
obs columns · 12
categorical · 8 categories
categorical · 6 categories
continuous · 201 – 5,942
continuous · 0.0 – 4.9 %
continuous · median 2,197
categorical · 2 categories
2,638 cells · 8 categories · 0 missing
- 01.obs is AnnData’s per-observation annotation table: one row per cell (n_obs), one column per label such as
cell_type,sample, orn_genes_by_counts. ObsInspector never touches your expression matrix (X) — only the labels beside it. For bulk and methylation data the rows are samples rather than cells.
Everything about a column, and a place to try questions.
Once the node has run, every column gets a summary: dtype, the number of unique values (NaN excluded), how many cells are missing, the most frequent values as a bar chart, summary statistics for numeric columns, and a sample of the actual values you’ll be grouping on.
Table View
The obs preview the node returned — 50 rows by default, up to 500. Search across text columns, sort by any column, hide the ones in the way.
Query
A pandas-style expression run against the upstream data: how many rows match, and a 200-row preview of them. Exploration only — it never subsets the workflow.
Column Ops
Drop, rename and convert dtype with multi-select pickers and a rename-pair editor. Apply saves the parameters; Apply & Run also re-runs the node.
Columns
6 columns · 2 selectedcell_typecategory5 uniqsamplecategory4 uniqn_genes_by_countsint644,210 uniqpct_counts_mtfloat649,994 uniqleidencategory11 uniqdoublet_scorefloat6410,000 uniq
cell_type
category
5
0
value_counts · top 5
How many values to keep per column is the node’s value_counts_top_n (10 by default, 1–100). Numeric columns show min / max / mean-style statistics instead of a bar chart.
Table View
search text columnssort · hide columnsshowing 4 of 50 preview rows| cell_id | cell_type | sample | n_genes_by_counts | pct_counts_mt |
|---|---|---|---|---|
| AAAC-1 | T cell | S1 | 2,431 | 4.2 |
| AAAG-1 | Monocyte | S1 | 1,988 | 6.7 |
| AACC-1 | B cell | S2 | 2,104 | 3.1 |
| AACT-1 | NK cell | S2 | 1,752 | 5.8 |
A query tab for “how many cells would that be?”
Type a pandas-style expression over the obs columns and press Run. The workspace evaluates it against the upstream node’s cached data and answers with the number of matching rows and a preview of up to 200 of them. It is exploration only: the workflow is not subset. When the expression is the one you want, put a Subset Data node after this one and use it there.
- Needs the upstream node to have run — the query reads its cached data, not a re-computation.
- A query can arrive pre-filled from an AI Chat conversation; edit it or Run to make it yours.
- Copy the expression with one click to reuse it in Subset Data or a script.
Example queries
- pct_counts_mt < 20
- n_genes_by_counts > 500 and total_counts < 50000
- leiden in ["0", "1", "2"]
- sample == "patient_a" and cell_type != "doublet"
Drop, rename, convert — and keep it reproducible.
Obs Inspector is for tidying labels, not rewriting data. The edits it allows are deliberately small, and each one is stored as a node parameter, so re-running the workflow reproduces exactly the same obs table. Targets that no longer exist are skipped and reported back rather than failing the run.
Drop
Remove columns you no longer need — leftover scores, scratch labels, stale batch keys — so the obs table stays legible downstream.
Rename
Give a column a human-readable name without a merge node or a scratch script. Pairs are edited one row at a time; the old key is replaced.
Convert dtype
Cast a column to category so plots and group-bys behave, or back to plain string when a label was over-eagerly categorized.
Applied on Run in a fixed order
- 1
drop - 2
rename - 3
to_category - 4
to_string
Recorded as node params
{
"drop_columns": ["doublet_score"],
"rename_pairs": [{ "old": "leiden", "new": "cluster" }],
"to_category": ["sample"],
"to_string": []
}- Apply saves the parameters; Apply & Run saves and re-runs the node.
- Deterministic — same params, same obs table.
- Auditable — the edit lives on the node, not in a lost notebook cell.
- Reversible — clear the param and re-run; the column comes back from upstream.
The node exposes 7 parameters in total; the ones above plus the preview and value-count limits. Full reference: Obs Inspector in the Input / Output nodes.
Sits beside QC, Figure Canvas, and Reports.
ObsInspector is one of Oreumics’ focused workspaces — a full-screen surface you open on top of the canvas for one job, then close. Use it right after loading, merging a sample sheet, or annotating, before you branch into clustering, DEG, or a report.
QC Workspace
Interactive thresholds, MAD recommendations, live re-rendering.
Figure Canvas
Free-layout figure plates — pin, arrange, annotate, export PNG.
Reports
HTML and PDF reports drafted from the finished workflow.
ObsInspector
You are hereBrowse, query and lightly edit obs columns — the workspace you are on.
Included with Beta Access
Read your obs table before you trust it.
Load an H5AD, drop in an Obs Inspector node, open the workspace, and get to know your cell metadata in a minute.