Skip to main content

Bayes Net Structure Learning from Uncertain Interventions

We show how to learn causal structure from data generated from systems subject to perturbations (interventions) with unknown effects. Our approach is to treat the interventions, as well as the other system variables, as random variables, and to learn a joint graph topology using Bayesian inference. This is in contrast to the standard approach, which assumes that the effects of interventions are known, and the graph structure is only learned on the system variables. We show that, on a datatset consisting of protein phosphorylation levels measured under various perturbations, learning the targets of intervention results in models that fit the data better than falsely assuming the interventions have known targets. Furthermore, learning the children of the intervention nodes is useful for such tasks as drug and disease target discovery, where we wish to distinguish direct effects from indirect effects. We illustrate the latter by correctly identifying known targets of genetic mutation in various forms of leukemia using microarray expression data. This is joint work with Daniel Eaton.