Bootstrapped Markov blanket discovery.

<p>Bayesian network inference (BNI) is performed on 400 bootstraps of size 20000. The -axis represents the fraction of bootstraps that a feature occurs in the Markov blanket of integration proximity in a resulting Bayesian network, i.e. the confidence we have in an edge. The -axis represents the mean conditional mutual information (CMI) of integration proximity with a feature across all Markov blankets in which this feature occurs, i.e. the strength of an edge. Note that features that do not occur in the Markov blanket of any bootstrap, i.e. are never considered relevant for integration proximity by the BNI approach, are not shown in this figure.