Figure 3
From: Discrete profile comparison using information bottleneck
Sequence logos for |C| = 20, 40, 80, showing several features of IB discretization. First, variable numbers of clusters are assigned to different amino acids according to their overall frequencies: A and L are more common, while C is least common. Second, clusters capture strongly- and weakly-conserved variants, as well as some chemical similarities: I, V, L, and M are all hydrophobic.