Cluster single-cell expression data and export marker evidence

Why use this

1. A single-cell researcher may receive a count matrix without a clear record of preprocessing, metadata alignment or analysis choices. Attractive cluster plots can distract from questions about input suitability and the decisions that produced them.

2. This job could make the proposed analysis and review requirements explicit before computation. Separating quality checks, clustering choices and tentative annotation helps an expert judge what evidence would be needed to interpret any later result.

Why an agent / simpler alternative

Useful judgment

Reconcile input manifests, expose ambiguous preprocessing and organize approved analysis choices without assigning unsupported cell identities.

Simpler alternative

Use an established lab pipeline when its assumptions and parameters already fit the dataset.

Unsuitable when

The workflow would use identifiable biological data without approval or present exploratory clusters as validated biological findings.

Proposed inputs

1. Name

   Count-data manifest

   Owner

   Authorized researcher

   Minimum scope

   Matrix format, raw-count status and sample/cell identifiers

   Quality check

   Check dimensions, naming and preprocessing provenance

2. Name

   Analysis brief

   Owner

   Domain lead

   Minimum scope

   Approved quality criteria, clustering choices and optional analyses

   Quality check

   Separate required steps from optional annotation

Access and setup — proposed or unknown

Proposed minimum

1. Inspect synthetic manifests and a proposed analysis checklist; no data upload, installation or scientific computation.

Compatibility unknowns

1. Current library versions, input support, compute needs and runtime compatibility are unknown.

External actions requiring approval

1. Sensitive-data access, external uploads, compute spending, optional analyses and output overwrites require authorization.

Expected deliverable and review checks

Proposed artifact

Analysis-plan candidate with input checks, parameter record, expected artifacts and expert-review questions.

Critical checks

1. Require identifier alignment and raw-count evidence; retain all parameter choices; distinguish marker evidence from cell-type interpretation.

Missing output evidence

This overlay adds no new output inspection, execution or validation; consult the referenced canonical source outcome limits.

Failure signals and proposed responses

1. Signal

   Input values have undocumented preprocessing

   Why it matters

   The planned analysis may not match its assumptions

   Proposed response

   Hold execution and obtain provenance.

2. Signal

   Cell metadata uses mismatched identifiers

   Why it matters

   Labels could be attached to the wrong observations

   Proposed response

   Reconcile IDs before any computation.

Effort and cost — unknown

Setup work

1. Document provenance and approved analysis decisions.

Ongoing review

1. Experts review quality findings, markers and any later annotations.

Unknowns

1. Dataset quality, memory demand, method suitability and computational costs are unknown.

First check — proposed, unexecuted

Status

proposed_unexecuted

Synthetic fixture

Invent a matrix manifest with mismatched cell IDs and unclear raw-count status; prepare a preflight checklist only.

Acceptance checks

1. Flag both input problems; leave clustering/annotation unperformed; make no biological claim.

Stop conditions

1. Stop if provenance, privacy scope or analysis authority is unresolved.

Limits

1. A manifest check cannot validate clusters, marker significance, biological identity or reproducibility of a future run.