Skip to main content
. 2021 Oct 28;38(3):597–603. doi: 10.1093/bioinformatics/btab746

Fig. 1.

Fig. 1.

(Left) Inputs to CpG Transformer. Cell, DNA and CpG embeddings are applied row-, column- and element-wise, respectively. (Middle) Illustration of sliding window row attention and (full) column attention. For sliding window row attention, every query attends to keys in the same row within a fixed window. For column attention, every query attends to the keys from all elements in the same column. (Right) A single CpG Transformer layer