| Title: | Evidence-Grounded Cell-Type Annotation for Seurat Objects |
|---|---|
| Description: | Annotates single-cell RNA-seq clusters by measuring curated marker panels in the data and letting a language model decide which identity those measurements support, with every claim resolving to a publication and its source sentence. Reads Seurat .rds objects natively and runs from raw counts, against one shared configuration, one set of prompts and one marker resource. |
| Authors: | Haoran Chen [aut, cre], Chen Cao [aut] |
| Maintainer: | Haoran Chen <chenhaoran@stu.njmu.edu.cn> |
| License: | file LICENSE |
| Version: | 0.1.0 |
| Built: | 2026-09-24 15:31:23 UTC |
| Source: | https://github.com/HaoRan-Code/scMarkerAgent |
Runs the full scMarkerAgent pipeline on one Seurat object: QC, normalization, Leiden clustering (always recomputed), genome-wide differential expression, candidate retrieval from the curated marker resource, the annotating agent and the evidence review that checks it, and the delivered result package (tables, audit sidecar, figures, offline viewer and ZIP).
annotate( input, tag = "dataset", species, tissue, counts_source, disease = "Normal", resource_dir = NULL, work_dir = NULL, clustering_resolution = NULL, offline = FALSE, cluster_annotation = TRUE )annotate( input, tag = "dataset", species, tissue, counts_source, disease = "Normal", resource_dir = NULL, work_dir = NULL, clustering_resolution = NULL, offline = FALSE, cluster_annotation = TRUE )
input |
Path to a Seurat '.rds' file carrying raw counts. |
tag |
Short name for the run; names the results directory. |
species |
'"Human"', '"Mouse"' or '"Rat"'. |
tissue |
Free-text tissue, resolved against UBERON. Several names joined by '"|"' (e.g. '"urinary bladder|adipose tissue"') are one operator-fixed list: the marker menu is the union of each listed tissue's own UBERON closure, every retrieved candidate records which listed tissue admitted it, and the list is written verbatim to the run manifest. A single name is the default behaviour. |
counts_source |
Where the raw counts live, as '"<assay>/<layer>"' (for example '"RNA/counts"'). Required and never guessed. |
disease |
Disease context; several values may be given as a character vector. Defaults to '"Normal"'. |
resource_dir |
The downloaded marker resource directory (see [download_resources()]). Verified before the run starts. |
work_dir |
Cache and results root; defaults to './.scmarkeragent'. |
clustering_resolution |
Leiden resolution; 'NULL' keeps the shared default. |
offline |
Run the deterministic stages only, without contacting a model. |
cluster_annotation |
Set 'FALSE' to skip the annotating agent; the label is then the top of the retrieval order, and every field recording how the label was produced says so. |
The parsed 'run_manifest.json', invisibly: run status, per-stage outcomes, and the paths of every delivered file.
Checks that the file exists and that its format belongs to this arm, and explains the way out when it does not. The check is at the entry point rather than inside the reader because the two failures it prevents look nothing alike to a user: an '.h5ad' handed to this arm dies inside 'readRDS' with a message about an unknown format, and a path typo dies even later, after preprocessing has already written a cache directory. Both are the same mistake, and both are answerable here in one sentence.
check_input(path, expected_arm = "r")check_input(path, expected_arm = "r")
path |
Path to the input object. |
expected_arm |
Which arm is asking; '"r"' for this package. |
The normalized path, invisibly usable as the validated input.
Download the curated marker resource
download_resources(dest, url = NULL)download_resources(dest, url = NULL)
dest |
Directory to unpack into; pass it to a run as 'resource_dir'. |
url |
Archive URL. Defaults to the one recorded in the resource index; when no URL is configured yet, the function explains how to install the bundle by hand. |
The destination path invisibly.
The shipped resource index
resource_index(path = .index_path())resource_index(path = .index_path())
path |
Optional path to an index file; defaults to the packaged one. |
The index as a list: bundle name, total size and per-file size and SHA-256.
Verify an existing resource directory
verify_resources(dest)verify_resources(dest)
dest |
Directory holding the unpacked resource. |
'TRUE' invisibly when every indexed file is present and intact; otherwise it stops with the list of problems.