GeneLLM / cfRNAraw sequence reader
Nature Communications · 2026

Raw sequence.
Readable signals.

GeneLLM is described as learning from plasma cfRNA sequences before they are reduced to annotated genes. This transparent lab makes that idea inspectable: bring a sequence, see its composition, keep the evidence.

AUG CUA
GGA UAC
CUA GGC3-mer windows · visible by design

Inspect a sequence

Paste RNA letters or use the sample. The analysis is local, reproducible, and educational. It is not a cancer test and does not reproduce GeneLLM or infer a diagnosis.

Only A, U, G, and C are counted. Other characters are removed and reported.
Grounded in the supplied post

“Decoding cancer circulating transcriptomic signatures with language models.” Nature Communications (2026), DOI 10.1038/s41467-026-74411-3. The paper context motivates the question; the calculations above are deliberately simpler and fully inspectable.

Literal RNA counts, overlapping windows and local exports

Read the explanation

The supplied sample contains twenty-six retained RNA letters: seven A, six U, seven G and six C. At thirty pixels per base the four bars are two hundred ten, one hundred eighty, two hundred ten and one hundred eighty pixels long. Together they represent twenty-six bases. The tool uppercases text and removes every character outside A, U, G and C; it does not align genes or run a language model. These are literal counts from supplied text, not patient measurements or a diagnosis. Sliding a three-letter window across twenty-six letters gives twenty-four positions, the length minus two. Bars compare twenty-six base positions with twenty-four overlapping windows at twenty pixels per count. Adjacent windows share two letters; they are not independent biological samples. The code groups identical three-letter strings, sorts by frequency then alphabetically, and displays only the five highest groups. It does not estimate a disease signature, train embeddings or reproduce the paper named in the source citation. Seven G plus six C equals thirteen out of twenty-six retained bases, or fifty percent. At twenty pixels per base the bars compare thirteen G-or-C bases with twenty-six total bases. Export JSON includes cleaned sequence, counts, GC percentage, top three-mers and removed-character count; CSV includes only length, GC percentage and the four counts. Editing without Analyze preserves the old export. Reload analyzes the default sample and overwrites saved prior results rather than restoring them. This educational computation has no patient classification or model inference.

Super generates helpful tools and automates fact-checking across the internet proactively. If you enjoyed this tool, build your own with Super and share it with a friend.