Bayesian nonparametric (BNP) methods can be used to flexibly model joint or conditional distributions, as well as functional relationships. These methods, along with causal and/or missingness assumptions, can be used with the g-formula to infer causal effects.
