Discovery of causal mechanisms from data is a fundamental problem of several computational disciplines including computer science, statistics, and applied mathematics. Obtaining data from randomized controlled experiments, while being fundamental for discovery of causality, is very expensive and can often be infeasible and unethical. On the other hand, non-randomized observational (e.g., case-control, case-series, time-series) data that is collected without experimental interference of the values of variables is highly abundant in the public domain and otherwise can be often collected cheaply. Over the last 20 years, many sound algorithms have been proposed that can leverage observational data to infer causal relations.

The Computational Causal Discovery Laboratory is contributing to this effort by developing, testing and applying causal discovery methods suitable for biomedical (clinical, molecular, and multi-modal) and general data of high-dimensionality. Specifically, we are interested in causal discovery methods/approaches that:

  • find direct causes and/ordirect effects of the response variable of interest
  • find a local causal pathway/neighborhood or causal graph around the response variable of interest
  • find causal paths between a pair of variables
  • reverse-engineer an entire causal network
  • predict the effect of manipulation of one variable on another
  • infer directionality of causation between pairs of variables from non-experimental data
  • relax standard assumptions to facilitate more accurate discoveries
  • optimally design randomized controlled experiments