Figure 5

Comparison of alternate methods of computing letter height within a stack. These examples were built using the profile HMM for Peptidase_C14 Pfam protein domain family (PF00656), built using the hmmbuild tool from HMMER 3.1. (A) Information content (all): letter stack height is the information content of the column, and all letters subdivide that stack height according to their probability. For models built using strong priors, as with HMMER 3.1, it is common to see an unreadable clutter of below-background letters at the bottom of the stack. See an example under the prominent D and Q at position 125. (B) Information content (above-background): a less noisy variant, in which stack height is also based on information content, and that height is divided only among letters with above-background probability. Notice the reduced letter clutter in position 125. (C) Score: a variant in which a letter’s height depends on the score of that letter at that position. Only positive-scoring letters (those with above-background probability) are included in the stack. In this case, the height of a stack does not have any inherent meaning – it is simply the sum of all letter heights. As an example of this, note that the stack for position 126 is slightly taller than the stack for position 125, despite the fact that position 125 is much more conserved as seen in Figure 5A and B. This is because the more conserved 125 has only two positive scoring letters (D=3.4, Q=2.6), while position 126 has five (C=2.0, A=1.8, S=1.7, T=0.8, M=0.04). The stacking order in Figure 5C may differ from the order in Figures A and B. This is because relative letter height in A and B depends only on the frequency in the distribution $\overrightarrow{\mathrm{p}}$, whereas letter height in 5C depends on the score, which accounts for the background distribution, s _i  := log ₂(p _i /q _i ).