Figure 2
From: Efficient pairwise RNA structure prediction and alignment using sequence alignment constraints
Comparison of parameters using two different training datasets. The parameter values of the LSU/SSU trained model are plotted on a log scale against the parameters determined using the published Rfam training set. The Rfam training set excludes the SSU family to avoid overlap between the training sets. Because the plot is roughly linear (with a slope of 1), the parameter values appear to be relatively robust with respect to the source of the training data. Interestingly, the two off diagonal elements refer to the probabilities that a GU or UG pairs remain unchanged across the alignment (ie. a GU pair aligns to a GU pair), which are higher in the ribosomal dataset.