The Gibbs sampling algorithm proceeds by starting from a random assignment of peaks into clusters (and particular assignments therein) and then repeatedly re-sampling the assignment for each peak with the various posteriors computed by ignoring the peak being assigned.
Figure 2B shows that the posterior means computed by INLA are very close to those computed using MCMC and close to the true value of h.
The first one is based on the posterior probability computed by the model and the other is based on the derivative of the ROC function of the model.
In addition, the similarity measure used to conduct the segmental DTW search (Kullback-Leibler divergence) fits very well the posterior probabilities computed in the feature extraction stage.
This indicates that the high-risk posterior probability computed by the model is a good prognostic factor of lung cancer survival.
Support for clades was assessed using 1000 bootstrap replicates and posterior probabilities computed with MrBayes.
The posterior probability computed by our algorithm additionally provides an estimator of how likely the classification is to be correct.
Clade support was assessed with posterior probabilities computed with MrBayes and non-parametric bootstrapping implemented in RAxML [ 48] based on 1000 replicates.
As such, those cases where a class is assigned with only marginal differences in the posterior probabilities computed for each class should be treated with more care and in-depth assessment before a reliable diagnosis is made.
We use a full collection of posterior samples when computing the link probability, and then compute the area under the ROC curve for each.
