Real-data case study#

This article connects the public SCimilarity tutorial dataset to the TurboCell Atlas benchmark pipeline and shows the complete input-to-output story.

What this case study contains#

The case study includes:

  • the input data description

  • query definitions

  • pipeline-stage summaries

  • candidate-versus-rerank comparisons

  • benchmark summaries

  • interpretation and limitations

Why this page matters#

The case-study page is where method discussion becomes concrete. It shows not only whether a scenario worked, but what data were used, how the query was constructed, what artifacts were generated, and how the result should be interpreted.

Key result#

The most favorable current result remains the IPF myofibroblast centroid scenario, where TurboQuant preserved the exact top-100 neighborhood while reducing candidate-layer memory substantially.

Figure#

Real-data benchmark summary