A multimodal foundation model for single cells

Decode every cell's expression in its own language.

LingoCell aligns gene expression, natural language and spatial context in a single 512-dimensional space — embed any cell, retrieve it by phrase, and read back what the model thinks it is.

Submit a Job What is LingoCell?
Latest · News & releases

News.

View all →
Release

v0.4 ships Stage 2 v5_final and the 11-tissue Visium HD index.

Stage 2 spatial training reaches NMI 0.71 against pathologist niche labels. New checkpoint best_phase2.pt available on the lab page.

Read release notes →
Preprint

Multimodal alignment improves zero-shot annotation in Visium HD.

Benchmarks across breast, lung, colon and tonsil show consistent gains over scGPT and Geneformer baselines.

Open on bioRxiv ↗
Event

External Miyabi quota opens for community reproducibility.

HGC opens a small allocation for community users to re-run Stage 1.5 and Stage 2 on their own splits.

Request access →