A Bayesian model selection approach for mediation analysis

Mediation analysis is a powerful tool for discovery of causal relationships. We describe a Bayesian model selection approach to mediation analysis that is implemented in our bmediatR software. Using simulations, we show that bmediatR performs as well or better than established methods including the Sobel test, while allowing greater flexibility in both model specification and in the types of inference that are possible. We applied bmediatR to genetic data from mice and human cell lines to demonstrate its ability to derive biologically meaningful findings. The Bayesian model selection framework is extensible to support a wide variety of mediation models.

Combining dimensions and features in similarity-based representations

This paper develops a new representational model of similarity data that combines continuous dimensions with discrete features. An algorithm capable of learning these representations is described, and a Bayesian model selection approach for choosing the appropriate number of dimensions and features is developed. The approach is demonstrated on a classic data set that considers the similarities between the numbers 0 through 9.