RUN

• Evolution of the log likelihood during training
  The type of curves you get depends on the chosen training scheme. When training with EM, the programs plots the log likelihood of the whole HMM against twice the number of EM cylcles. After each E-step, the log likelihood ist plotted after each of two partial M-steps: after optimizing the recombination probability, and after optimizing the branch lengths. When training with CML, the program plots the constrained log likelihoods, for each of the possible trees in turn, against the number of gradient ascent steps. In both cases, this provides an easy way to check if the training process has converged, that is, if the training parameters have been chosen appropriately.

• Plotting the phylogenetic trees
  The program draws the phylogenetic trees for the three possible topologies.

• Plot of the posterior probabilities and classification scores
  This plots a figure composed of six sub-figures. The sub-figures in the top row plot the posterior probabilities P(S_t|D), conditional on the data D, for the three states S_t in {1,2,3}, against the position t in the multiple alignment. The histograms in the bottom row show you the classification scores for three different regions. These regions are called No Rec (no recombination), Rec 1 (1st recombination), and Rec 2 (2nd recombination). This is motivated by the synthetic data and the Neisseria sequence. If you use your own data and want to change this, you need to modify the code in PlotRecombi.m. For the synthetic data, the correct topology in Rec 1 is State-2, the correct topology in Rec 2 is State-3. For the Neisseria data , the correct topology in Rec 1 is believed to be State-3, the topolgy in Rec 2 is unknown. In both cases, the dominant topology in region No Rec is State-1. You can get two different prediction from the program. The first classification is obtained from the Viterbi path, that is, the joint mode of P(S_1,...,S_N|D). You are prompted if you are interested in the single-site mode as well. If you give an affirmative response, the program plots a second histogram based on the mode of P(S_t|D).