Figure 7
From: Learning sparse models for a dynamic Bayesian network classifier of protein secondary structure
The learned dlink values for local and non-local positions in β strands. (A)The figure shows the dlink values learned from the real data for local correlations in β strands. The first plot illustrates the dlink values for the correlations between the ith position and the positions at i - L AA , ..., i - 1. The second plot depicts the dlink values for the correlations within the ith position. The third plot shows the dlink values for the correlations between the ith position and the positions at i + 1, ..., i + L AA . In each plot, rows represent the 20 amino acids for the observation vector at position i.(B) The figure depicts the dlink values learned from the real data for non-local correlations in β strands such that the β strand residues at positions i and j are known to make a bridge interaction. The first plot illustrates the dlink values for the correlations between the ith position and the positions at j - L AA , ..., j - 1. The second plot depicts the dlink values for the correlations between the ith position and the jth position. The third plot shows the dlink values for the correlations between the ith position and the positions at j + 1, ..., j + L AA . In each plot, rows represent the 20 amino acids for the observation vector at position i. In this experiment, L AA is chosen as 5 corresponding to a window of 11 residues. In both figures, gray colored bins represent dlinks that are not present in the resulting sparse model.