How this works

Every clade at the chosen rank is compared with its whole domain. At each position, the clade's distribution over A/U/G/C/deletion is measured against the domain's distribution by Jensen–Shannon divergence (JSD, in bits): 0 means the clade looks exactly like the domain there, and 1 means they share no state at all. For several positions the per-position values are averaged, so a high score means the clade is unusual across the region, not at a single site.

Counts use one sequence per species: the top rRNA hit in its representative genome. Ambiguity codes are split across the bases they allow, missing sequence (~) and calls below the PP threshold are left out, and a deletion counts as a fifth state, so a clade that lost a base everyone else has scores high.

The PTC and decoding-center regions are 3D shells from the E. coli 70S structure 7K00. They include every nucleotide with any heavy atom within the radius of any heavy atom of the center, as in "Find positions near…" on the tree explorer.