scMarkerAgent

scMarkerAgent: code + marker resource Everything needed to run scMarkerAgent: the Python package for .h5ad data, the R package for Seurat .rds data, and the curated marker resource used for annotation. The source code is maintained at: https://github.com/HaoRan-Code/scMarkerAgent The GitHub repository contains the code only because the curated resource is too large for Git. An exact copy of the corresponding code tree is also archived here as scmarkeragent-code.tar.gz. Contents scmarkeragent-code.tar.gz — 3.0 MiBContains the Python and R packages, the resource download contract in required_files/, and the README walkthrough. Unpacks to scmarkeragent-code/. scmarkeragent-curated.tar.gz — 148 MiBContains the curated marker resource (approximately 760 MiB unpacked), including: the marker table; the source-sentence table, aligned row-for-row with the marker table; the Uberon ontology; ortholog tables; and resource_manifest.json, which records the size, checksum, and licence of each resource file. Unpacks to scmarkeragent-curated/. scmarkeragent-curated.SHA256SUMSContains SHA-256 checksums for every file in the curated resource directory. Verify the resource After downloading the archive, unpack it and verify its contents with: tar -xzf scmarkeragent-curated.tar.gz sha256sum -c scmarkeragent-curated.SHA256SUMS Either package can also download the resource automatically: scmarkeragent download-resources or, in R: scmarkeragent::download_resources() Both implementations verify every downloaded file against the bundled SHA-256 index. A resource directory unpacked from this archive is therefore interchangeable with one downloaded by the package. Install and run Python For .h5ad input. Python 3.12 is supported. tar -xzf scmarkeragent-code.tar.gz pip install ./scmarkeragent-code/python scmarkeragent verify-resources --resource-dir scmarkeragent-curated scmarkeragent annotate --input your.h5ad --tag run1 --species Human --tissue blood --counts-source X --resource-dir scmarkeragent-curated R For Seurat .rds input. Requires R >= 4.2. The R package, together with presto—the differential-expression engine used by scMarkerAgent and not available from CRAN—is distributed through the project's CRAN-like R-universe repository. The source under scmarkeragent-code/R in this archive is identical to the distributed package source. options(repos = c( scmarkeragent = "https://haoran-code.r-universe.dev", CRAN = "https://cloud.r-project.org" )) install.packages("scmarkeragent") scmarkeragent::verify_resources("scmarkeragent-curated") scmarkeragent::annotate( input = "your.rds", tag = "run1", species = "Human", tissue = "blood", counts_source = "RNA/counts", resource_dir = "scmarkeragent-curated" ) LLM endpoint and offline mode The annotation agent and the evidence review that checks it use an OpenAI-compatible API endpoint. Set: OPENAI_API_KEY and, when required by the provider: OPENAI_BASE_URL The --offline option in Python, or offline = TRUE in R, runs only the deterministic stages and does not call the LLM endpoint. For step-by-step setup instructions, configuration options, and output descriptions, see: scmarkeragent-code/README.md Licence Code: PolyForm Noncommercial 1.0.0. The licence text is included as LICENSE inside scmarkeragent-code.tar.gz. Curated marker resource: CC BY-NC 4.0. Source sentences in the marker resource are quoted from PubMed Central open-access articles, and each row records its corresponding PMCID and/or PMID. The ontology and ortholog files retain their respective third-party terms. File-specific provenance and licence information is recorded in resource_manifest.json.