robin workflow¶
Run the Little John orchestrated pipeline on BAM files under a watched directory: preprocessing, optional BED conversion, analyses (MGMT, CNV, target, fusion), classifiers (Sturgeon, NanoDX, PanNanoDX, random forest), and optional NiceGUI monitoring.
Startup sequence
When you run this command, ROBIN checks models, optionally validates --reference, asks you to type I agree to the research disclaimer, then (with the GUI enabled) may prompt for the GUI password. On the default Ray path, the browser UI is only started if --work-dir is set. Details: What happens at startup.
Synopsis¶
| Argument / option | Required | Description |
|---|---|---|
PATH | Yes | Directory containing (or receiving) BAM files. Must exist. |
-w / --workflow | Yes | Comma-separated job types or legacy queue:job steps (see Job types). |
--center | Yes | Site or study label (e.g. Sherwood, Auckland) — used in outputs and reports. |
--target-panel | Yes | Panel name: built-in (rCNS2, AML, PanCan, …) or custom from robin add-panel. Run robin utils sequencing-files --help to see choices on your install. |
-d / --work-dir | No | Base directory for all run outputs. |
-r / --reference | No* | Path to reference FASTA. If provided, ROBIN validates the file and ensures an index (e.g. .fai). Required for analyses that need a reference. |
* Omit only if your chosen workflow truly does not need a reference; most real pipelines pass --reference.
Workflow string formats¶
Simplified (recommended)¶
Comma-separated job types only — queues are assigned automatically:
Behaviour:
preprocessingis always inserted as the first step if missing.bed_conversionis inserted before classification jobs when needed (sturgeon,nanodx,pannanodx,random_forest).
Legacy (explicit queues)¶
Use queue:job steps, e.g.:
Execution engine¶
| Mode | Flag | Notes |
|---|---|---|
| Ray (default) | --use-ray | Distributed task execution; optional Ray dashboard (--ray-dashboard / --no-ray-dashboard). |
| Threading | --no-use-ray | Falls back to threaded workers; analysis worker counts apply per queue mode. |
Presets (--preset): p2i, standard (default), high — adjust Ray actor grouping and CPU caps for different hardware (e.g. P2i vs server).
GUI (NiceGUI)¶
| Option | Default | Description |
|---|---|---|
--with-gui / --no-gui | GUI on | Launch workflow monitor in the browser. |
--gui-host | 0.0.0.0 | Bind address. |
--gui-port | 8081 | Port for the GUI. |
Logging and progress¶
| Option | Description |
|---|---|
--log-level | Global level: DEBUG, INFO, WARNING, ERROR (default ERROR). |
--job-log-level | Repeatable, e.g. preprocessing:DEBUG, mgmt:WARNING. |
--verbose / -v | Verbose CLI and traces. |
--no-progress | Disable file progress bars. |
File handling¶
| Option | Description |
|---|---|
--no-process-existing | Only process BAMs that appear after startup (skip files already on disk). |
--no-watch | Do not watch the directory for new files (advanced; default is to watch). |
Ray tuning¶
| Option | Description |
|---|---|
--ray-num-cpus | Cap CPUs for Ray (auto if omitted). |
--queue-priority | Repeatable, e.g. preprocessing:10, mgmt:5 (Ray mode). |
--show-priorities | Print queue priorities and exit. |
--analysis-workers | Workers per analysis queue (default from code constant, often 1). |
--preprocessing-workers, --bed-workers | Workers for preprocessing and bed_conversion. |
--legacy-analysis-queue | Single shared analysis queue instead of per-type queues (threading path). |
Deduplication¶
--deduplicate-jobs (repeatable) — job types that should run at most once per sample even when multiple triggers fire (e.g. sturgeon, mgmt).
Custom per-job shell commands¶
--commands / -c (repeatable) — optional mappings job_type:shell_command for custom steps (advanced; see robin workflow --help).
Inputs and sequencing assumptions¶
ROBIN expects real-time BAMs with ≤ 50,000 reads per file and read-count–based rollover in MinKNOW. See the README.
Exit¶
Stop with Ctrl+C — the runner attempts graceful shutdown; complex runs may not always exit instantly.