DNA representation, explained.
Text boundaries and biological reading frames are different. This illustration explains the distinction; it does not demonstrate Helicase’s tokenizer or accuracy.
- AT
- G
- TC
- CTtwo codons
- GG
- GCC
- G
- CT
- CT
- G
- CT
- CGtwo codons
- GC
- CT
- T
- CT
- GGtwo codons
- CC
- GT
- G
Across the whole gene, 220 of 796 tokens mix pieces of two codons (dashed). That describes the grouping, not an error in the DNA or a model’s accuracy. The letters are unchanged.
- ATGMet0
- TCCSer3
- TGGTrp6
- GCCAla9
- GCTAla12
- CTGLeu15
- CTCLeu18
- GGCGly21
- CTTLeu24
- CTGLeu27
- GCCAla30
- GTGVal33
Each group shows three bases, their amino acid and their position (counting from zero). The full sequence has 502 codons. Sequence tools compute this readout from a known frame; these are not Helicase model predictions.
Illustrative. Source: rat prostacyclin synthase (NM_031557). Tokenised with the 34 of Qwen3’s 151,387 merges that join only A, C, G and T; the final displayed token is clipped to the shared window. GeneLoop uses sequence tools today. Helicase reads DNA at single-base resolution; this illustration is a teaching example, not a measurement of it. View the source sequence