Explore single-cell and single-nucleus data in context.

Prepared matrices for cells and nuclei

Start with AnnData (.h5ad) or supported 10x gene-expression matrices (.h5 or Matrix Market with barcode and feature files). The same matrix-based tools can explore cells or nuclei; retain sample, donor and assay metadata so the origin of each observation stays clear.

Generate the expression matrix upstream with settings appropriate to your assay. Raw-read alignment and counting are outside this workflow. For 10x data, review the Cell Ranger guidance for single-nucleus RNA-seq when preparing the input.

Inspect QC distributions and choose thresholds for the tissue, assay and cells or nuclei you measured. Record those choices before filtering. Matrix-format compatibility does not establish a separately validated single-nucleus pipeline.

Start with data you can inspect

Before interpreting a visualization, inspect the dataset's dimensions, identifiers and available metadata. Know which observations and features are present, what preprocessing has already happened and which values are raw or transformed.

The current software acceptance checks use synthetic fixtures to test imports, selections, metadata access and h5ad export/reload. This is evidence about software behavior, not a guarantee that every dataset or upstream pipeline has been qualified.

Use UMAP as an exploration view

A UMAP view can help you explore patterns, select observations and inspect related metadata. Keep the embedding connected to the underlying cell or nucleus identifiers so a selection remains meaningful when you inspect a table or use Python.

An embedding is a view of processed data. Interpretation depends on input features, preprocessing and method settings. Record those choices and examine the underlying data before treating a visual group as a biological finding.

Review annotation candidates in context

Use annotation candidates as inputs to review alongside marker evidence, sample context and the limitations of the method or reference used. The availability of an annotation tool is not evidence that its labels have been validated for your dataset.

Numbers of cells or nuclei do not substitute for independent biological replication. Retain donor and sample metadata so that exploration does not erase the design of the experiment.

Move between figures, tables and Python

Selections should help you examine data rather than become a detached picture. Inspect selected observations, examine variables in the notebook and export data with the metadata needed to understand it.

Before relying on a workflow, test a small export and reload it. Compare identifiers, dimensions and the values that matter for your analysis. Keep a copy of the original dataset and save the notebook with the analysis settings.

An example you can explore now

The homepage's butterfly-to-data journey includes 240 simulated cells and 16 features. You can inspect a cell, download the synthetic counts and save a figure. The illustration and data are conceptual examples, not measurements of the butterfly.

For desktop testing, read the beta notes and download the current Windows installer from the release section. Learn more about local analysis and optional AI before bringing sensitive research into any workflow.