Genome is coded by sequences of nucleotides. Turns out sequences of tokens are exactly what modern ML excels at. Adapting models thought for text to genomics has been a tremendous accelerator to the field.

For instance, DNA is usually cut into k-mers: fixed-length windows of k nucleotides. One popular choice of token size is 3-mers, and it happens to match codons: the exact triplets that map to amino acids.

6-merization of a 12-nucleotide sequence: jump the window by k for non-overlapping tokens, or slide it by 1 to get overlapping tokens.