Who uses single-cell AI models? Geneformer, scGPT, scVI and more, by preprints

1,247 preprints read so far say they used a single-cell AI model, scVI most of all; 42% of them also ran single-cell experiments themselves. Updated daily (as of Nov 16)

Counts preprints first posted Apr 2025–Nov 2026; Nov 2026 is still being read.

scVI is the single-cell AI model used in the most preprints.

1,247 preprints read so far say they used a single-cell AI model; the count is preprints using each.

ModelPreprints

42% of preprints using a single-cell AI model made their own single-cell data.

525 of the 1,247 preprints that used a single-cell AI model also ran single-cell or spatial experiments themselves; the rest did not report single-cell experiments of their own.

Which platforms do model users generate data on?

The single-cell and spatial platforms behind the data, for preprints that used a single-cell AI model on single-cell or spatial data they made themselves. Chromium leads: 356 of the 438 that used a single-cell or spatial platform we track. A preprint can use more than one, so the shares add to more than 100%. These counts grow as more preprints are processed.

PlatformPreprints (share)
  • Chromium10x Genomics · Droplet single-cell356 (81%)
  • Xenium10x Genomics · In situ imaging68 (16%)
  • Visium10x Genomics · Spatial sequencing, incl. Visium HD67 (15%)
  • EvercodeParse (QIAGEN) · Split-pool single-cell24 (5%)
  • CosMxBruker (NanoString) · In situ imaging17 (4%)
  • PhenoCyclerAkoya (Quanterix) · Multiplex protein imaging, formerly CODEX13 (3%)
  • MERSCOPEVizgen · In situ imaging11 (3%)
  • RhapsodyBD (Waters) · Microwell single-cell5 (1%)

What other software do they use, and which public data?

Other software they usePreprints
Public datasetsPreprints

Fields: Bioinformatics (580), Cancer Biology (118), Genomics (97), Immunology (92), Neuroscience (78).

How does Tnuki know so much?

  1. Preprints analyzed dailyNew bioRxiv and medRxiv preprints are analyzed daily.
  2. Whole paper readWe assess the full text, tables, and supplementary files.
  3. Only real uses countedWe read to make sure the lab actually used the product, not just cited.
How we count

An AI model reads each bioRxiv and medRxiv preprint and lists what its authors used — instruments, kits, services, software, reagents — and how: used in this work, someone else's data, or only mentioned. A preprint counts here once, when the model says the lab used it, if it was first posted in Apr 2025 or later. Names are merged by a second model (versions and spellings into one tool, brands under their owner). We show counts and shares, and the titles of the latest preprints, each linking to the preprint on bioRxiv or medRxiv; we show no other text from a preprint and no author names. Updated daily, most recently on Nov 16, 2026.

Want to reach out to the labs using single-cell AI models?

Get the PIs behind the preprints: institutions, labs, and how to reach them.