Figure 4

Comparison of alternate methods of producing an alignment logo from an input sequence alignment. Logos were built using the Rfam family seed alignment of 45 7SK sequences (RF00100, Rfam 11.0 [26]). Skylign performs necessary counting, weighting, and mixing by explicitly building a profile HMM using the HMMER tool hmmbuild. (A) Logo for the alignment based on observed frequencies (‘hmmbuild ‒‒symfrac 0 ‒‒wnone ‒‒enone ‒‒pnone’). Positions 105 and 106 highlight the fact that Skylign creates a logo position for every non-empty column in the alignment. Occupancy is ~2% because only one of 45 sequences contains a letter at each of those positions. Stacks are tall because stack height depends on observed counts, and there is no variability at these positions. (B) Logo for the alignment after applying sequence weighting [11] to account for sequence redundancy (‘hmmbuild ‒‒symfrac 0 ‒‒pnone’). The letter G at the first visible position of the weighted logo indicates much less conservation than does the G for the logo based on observed counts. This is because most of the support for high conservation of G at that position comes from a large set of highly similar sequences, and the importance of such redundant sequences is diminished under sequence weighting. Sequence weighting can also alter the represented occupancy rates, for example showing a weighted 7% occupancy for positions 105 and 106. (C) Logo for the alignment after applying sequence weighting, absolute weighting [12], and Dirichlet priors. This amounts to building a profile HMM under default HMMER conditions, except that a match state is created for every non-empty column in the alignment (‘hmmbuild ‒‒symfrac 0’). In the case of low weighted counts, as in positions 105 and 106, HMMER’s priors typically increase letter variance, leading to lower information content. (D) Logo for the HMM built using default ‘hmmbuild’, in which logo positions are created only for consensus, resulting in removal of positions 105 and 106.