Each human is made up of trillions of cells, every with its personal operate, whether or not it’s carrying oxygen, preventing infections, or constructing organs. Even inside the identical tissue, no two cells are precisely alike. Single-cell RNA sequencing (scRNA-seq) permits us to measure the gene expression of particular person cells, revealing what every cell is doing at a given second.
However there’s a catch: single-cell knowledge are large, high-dimensional, and arduous to interpret. Every cell may be represented by 1000’s of numbers — its gene expression measurements — which historically require specialised instruments and fashions to research. This makes single-cell evaluation sluggish, tough to scale, and restricted to skilled customers.
What if we may flip these 1000’s of numbers into language that people and language fashions can perceive? That’s, what if we may ask a cell the way it’s feeling, what it’s doing, or the way it would possibly reply to a drug or illness — and get a solution again in plain English? From particular person cells to whole tissues, understanding organic methods at this degree may rework how we examine, diagnose, and deal with illness.
At present in “Scaling Massive Language Fashions for Subsequent-Technology Single-Cell Evaluation“, we’re excited to introduce Cell2Sentence-Scale (C2S-Scale), a household of highly effective, open-source giant language fashions (LLMs) educated to “learn” and “write” organic knowledge on the single-cell degree. On this put up, we’ll stroll by way of the fundamentals of single-cell biology, how we rework cells into sequences of phrases, and the way C2S-Scale opens up new potentialities for organic discovery.