Source: https://docs.nygen.io/CyteArc/ Revision: fccee7b6e0c167e47ce605c8d292fba5b567ebfe Structured page: https://docs.nygen.io/CyteArc/index.json Out-of-core single-cell analysis CyteArc High-performance single-cell analysis where your data live. Open a published atlas, plot marker expression, and save your own annotations. Counts stay in object storage; your results live in a store you control. CyteArc supports RNA, ATAC, and multimodal analysis on your computer, an HPC cluster, or cloud compute. Benchmarked on datasets with up to 100 million cells. Try remote dataset mounting · Quick start · Explore the API One shared atlas, many analyses A published atlas can hold counts, prepared reference models, and saved analysis results. CyteArc lets you use each of these directly. Mounting creates a writable analysis store with its own metadata and results, while the count data stay in the shared source. Reanalyze shared data. Choose cells and run a new analysis, from feature selection and PCA to graphs, clusters, and markers. Counts stream from the source; your results are written separately. See remote dataset mounting . Map and label new cells. Reuse a prepared RNA reference’s genes, PCA model, and neighbour index to place query cells and transfer labels. Keep the query analysis separate from the reference. See Mapping cells and transferring labels . Explore before recomputing. Inspect saved embeddings and annotations, select populations, and read gene expression for the question at hand. Try the 1.1-million-cell Tabula Sapiens atlas , then Core plotting features . The available starting points depend on what the atlas contains: mapping needs a prepared reference, while exploration can start from published annotations and embeddings. Built for large cell atlases CyteArc streams count matrices in bounded blocks and stores RNA counts in both cell-wise and gene-wise layouts so each operation can read in the direction it needs. Memory and worker settings let you fit the work to your machine. Graphs, neighbour indexes, and other intermediate results still need memory and storage; the count matrix does not have to fit in RAM. Completed steps are saved and reused when their inputs and settings match. Change a clustering choice or examine another analysis branch without rebuilding every upstream result. See Reuse results and compare analysis settings for a worked example and Why CyteArc records provenance for how results record their inputs. Memory and execution explains resource settings and how to measure runtime and memory on your workload. Get your first result Start from remote data: mount Tabula Sapiens , plot its saved UMAP and a marker gene, then save a cell selection locally. No credentials are needed. Try CyteArc: the Quick start opens a prepared analysis of 100,000 cells sampled from the same atlas. For a smaller example that works through the analysis step by step, follow the PBMC walkthrough . Bring your own counts: Import and export covers supported formats, and the RNA-seq workflow shows the standard pipeline. Review Quality control across assays before interpreting the results. Use another assay: follow Interpret cell populations with scATAC-seq , Read RNA and protein together with CITE-seq , or Add a third modality with TEA-seq . Start with Installation to set up your environment, then follow the Quick start . For importing an existing analysis or translating individual steps, see the Scanpy guide or Seurat guide . Methods and help Methods and limitations describes the supported workflows, validation evidence, and scientific limitations. For help, check the FAQ and Glossary , ask a question , or report a bug . See Contributing to contribute code or documentation. The documentation index for LLMs lists guides and API references for assistants that read this site.