Fig. 6
From: A k-mer grammar analysis to uncover maize regulatory architecture
Local Sequence Context Defines Distinctive Groups ofk-mers Between Regulatory and Random Regions. a Schematic of aligning flanking contexts versus contrasting local sequence composition, as implemented in the “vector-k-mer” models in which k-mers that share a similar context would be represented by close vectors (vk-mers) in a geometric space. b The vector space obtained from core promoters (Vregulatory) and their corresponding control (Vrandom) define two different groups of closest k-mers (cosine similarity) to the ’CTATATA’ vector (vCTATATA). The group of closest k-mers in Vregulatory, when compared to the group formed in Vrandom, are more similar in sequence (shorter edit distance), and have in average more positive k-mer scores from the equivalent “bag-of-k-mers” model. This implies a semantic-like relationship between those k-mers in regulatory sequences versus random regions. c The group of k-mers closest in the Vregulatory space have similar positional preferences (blue solid lines) to CTATATA (black dotted line) in the region expected for the TATA element. d In addition, the group of k-mers closest in the Vrandom (red solid lines) do not show similar positional constraints to CTATATA (black dotted line) do not show positional preferences relative to the TSSs