Figure 12

Corrected CRF for the apparently signal-less polyadenylation signals. After re-alignment, we can see that the “non signal” polyadenylation signals actually do have a strongly conserved pattern of residues, but that unlike the “signal” motif, the motif also possesses significant dependencies. Notably, rather than a pair of A residues followed by a single T (as seen in the “signal” motif), these sequences possess a single A residue, followed with almost equal probability by an A or a T residue. That A or T residue strongly influences the identity of the subsequent residue—if the first is a T, then the second is also a T, if the first is an A, then the second is also an A. This variable pair of residues is then followed by, as in the “signal” motif, a trio of A residues ending the motif. Several other dependencies also show up. The interdependencies visualized here, are why PSSM and HMM models have failed to identify an alignment in, or adequately model this “non signal” signaling motif.