Abstract

Discovering causal effects is at the core of scientific investigation but remains challenging when mostly observational data is available. In essence, causal discovery infers cause-effect relations from specific correlation patterns involving at least three variables, which goes beyond the popular notion that pairwise correlation does not imply causation. Yet, in practice, causal representations have been difficult to learn and interpret, in particular, for high dimensional data such as state-of-the-art biological and biomedical data.

In this talk, I will outline some network reconstruction methods and a broad range of applications. In particular, our group has developed novel causal discovery methods and tools (i.e. MIIC , CausalXtract, MIIC -sdg, CausalCCC, MIIC seach&score) to learn cause-effect relationships in a variety of biological or biomedical data, from single-cell transcriptomics and live-cell imaging data to clinical data from medical records of patients. These Machine Learning methods combine multivariate information analysis with interpretable graphical models and outperform other methods on a broad range of benchmarks, in particular on complex non-linear datasets, while allowing for unobserved latent variables, that are ubiquitous in biomedical applications.