Cell2Sentence

Paste one cell's gene counts, get a cell-type call with the marker evidence behind it, plus the C2S sentence and reversible rank-to-expression transform. Runs entirely in your browser.

vandijklab/cell2sentence
gene, count per line
26 of 26
PREDICTED CELL TYPE
CD8+ T cell67%
CD3D · r1CD8A · r2CD8B · r3NKG7 · r5CCL5 · r6IL32 · r7CD2 · r8LCK · r9CD3E · r10TRAC · r11
Candidate types
CD8+ T cell
67%
NK cell
7%
CD4+ T cell
7%
Naive T cell
5%
Plasmacytoid DC
2%
B cell
1%

Rank-based marker enrichment against canonical panels, in the browser. Each panel marker found in the cell adds 1/log₂(rank+1), so top-ranked markers count most; scores are normalized by panel size and softmaxed. This is annotation, not the C2S LLM, but it answers the same question with visible evidence.

Cell-type calls use canonical marker panels scored by expression rank, so they are transparent and offline but only as good as the panels (human symbols, PBMC/immune coverage is strongest). The C2S transformer models give stronger calls on rare or ambiguous types but need Python and a GPU.
Cell2Sentence Playground — Gene Counts to Cell Type | Bibby | Bibby AI