Mistnet is an R package that produces probability densities over multivariate outcomes. Ecologists can use it to define probability densities over possible species assemblages.
Note that this software is under active development and that there will be major changes to the user interface and default methods in the next few months. Development should stabilize some time this summer (2014).
Mistnet models are stochastic neural networks, meaning that they include stochastic latent variables (like random effects) that account for correlations among the outcome variables that cannot be explained by the inputs.
The model uses a Generalized Expectation Maximization approach to model fitting (maximized penalized likelihood), as described in this paper by Tang and Salakhutdinov at ICML 2013.