Figure 11
From: Addressing the unmet need for visualizing conditional random fields in biological data
Visualized implied CRF from a misaligned polyadenylation signal model. Visualizing the “non signal” sequence regions, we see a strong disposition towards A and T bases in the marginal distributions, but no overwhelming pattern. PSSM and HMM methods fail to identify a pattern in these sequences, however, the fact that we do not restrict StickWRLD to only sequentially adjacent dependencies lets us see that there is a curious “echoing” pattern of dependencies between T15 and A17, T16 and A18, T17 and A19, and T18 and A20. This echo extends further, at lower residual and significance thresholds. This echo is a highly suggestive fingerprint of a misalignment in the data. We are seeing an interdependency between a T and an A base, 2 bases apart, occurring in several subsets of the input data, each shifted slightly from the next. Using StickWRLD to interact with and realign the data, we arrive at Figure 12.